Unified, high-performance TUI to index and search your local coding agent history. Aggregates sessions from Codex, Claude Code, Gemini CLI, Cline, OpenCode, Amp, Cursor, ChatGPT, Aider, Pi-Agent, and Factory (Droid) into a single, searchable timeline.

curl -fsSL https://raw.githubusercontent.com/Dicklesworthstone/coding_agent_session_search/main/install.sh \
  | bash -s -- --easy-mode --verify

# Windows (PowerShell)
irm https://raw.githubusercontent.com/Dicklesworthstone/coding_agent_session_search/main/install.ps1 | iex
install.ps1 -EasyMode -Verify

Search Results Across All Your Agents

Three-pane layout: filter bar, results list with color-coded agents (Claude, Codex, Gemini, Pi-Agent, etc.), and syntax-highlighted detail preview

---

Rich Conversation Detail View

Full conversation rendering with markdown formatting, code blocks, headers, and structured content

---

Quick Start & Keyboard Reference

Built-in help screen (press F1 or ?) with all shortcuts, filters, modes, and navigation tips

AI coding agents are transforming how we write software. Claude Code, Codex, Cursor, Copilot, Aider, Pi-Agent; each creates a trail of conversations, debugging sessions, and problem-solving attempts. But this wealth of knowledge is scattered and unsearchable:

• Fragmented storage: Each agent stores data differently—JSONL files, SQLite databases, markdown logs, proprietary JSON formats
• No cross-agent visibility: Solutions discovered in Cursor are invisible when you're using Claude Code
• Lost context: That brilliant debugging session from two weeks ago? Good luck finding it by scrolling through files
• No semantic search by default: File-based grep doesn't understand natural language queries; cass can add optional local ML search when model files are installed

cass treats your coding agent history as a unified knowledge base. It:

1. Normalizes disparate formats into a common schema
2. Indexes everything with a purpose-built full-text search engine
3. Surfaces relevant past conversations in milliseconds
4. Respects your privacy—everything stays local, nothing phones home

• Individual developers: Find that solution you know you've seen before
• Teams: Share institutional knowledge across different tool preferences
• AI agents themselves: Let your current agent learn from all your past agents (via robot mode)
• Power users: Build workflows that leverage your complete coding history

• "Search-as-you-type": Results update instantly with every keystroke.
• Edge N-Gram Indexing: We frontload the work by pre-computing prefix matches (e.g., "cal" -> "calculate") during indexing, trading disk space for O(1) lookup speed at query time.
• Smart Tokenization: Handles snake_case ("my_var" matches "my" and "var"), hyphenated terms, and code symbols (c++, foo.bar) correctly.
• Zero-Stall Updates: The background indexer commits changes atomically; reader.reload() ensures new messages appear in the search bar immediately without restarting.

• Local-only: Uses a MiniLM model via FastEmbed; no cloud calls.
• Manual install: Place model files under your data directory (see below); cass will not auto-download.
• Required files (all must be present):
  • model.onnx
  • tokenizer.json
  • config.json
  • special_tokens_map.json
  • tokenizer_config.json
• Vector index: Stored as vector_index/index-minilm-384.cvvi in the data directory.

When ML model files are not installed, cass uses a deterministic hash-based embedder as a fallback. While not "truly" semantic (it captures lexical overlap rather than meaning), it provides useful functionality:

Algorithm:

1. Tokenize: Lowercase, split on non-alphanumeric, filter tokens <2 characters
2. Hash: Apply FNV-1a to each token
3. Project: Use hash to determine dimension index and sign (+1 or -1) in a 384-dimensional vector
4. Normalize: L2 normalize to unit length for cosine similarity

When to Use:

• Quick setup without downloading model files
• Environments where ML inference overhead is unwanted
• Fallback when ML model fails to load

Override: Set CASS_SEMANTIC_EMBEDDER=hash to force hash mode even when ML model is available.

cass uses a custom binary format (.cvvi - Cass Vector Index) for storing semantic embeddings:

┌─────────────────────────────────────────────────┐
│ Header (32 bytes)                               │
│ ├─ Magic: "CVVI" (4 bytes)                      │
│ ├─ Version: u8                                   │
│ ├─ Precision: F32 or F16 (1 byte)               │
│ ├─ Dimension: u16 (e.g., 384)                   │
│ ├─ Entry count: u64                             │
│ └─ CRC32 checksum                               │
├─────────────────────────────────────────────────┤
│ Entries                                         │
│ ├─ content_hash: [u8; 32] (SHA-256)             │
│ ├─ source_id: varint                            │
│ ├─ agent: u8 (enum)                             │
│ ├─ timestamp: i64                               │
│ └─ vector: [f32|f16; dimension]                 │
└─────────────────────────────────────────────────┘

Features:

• Precision Options: F32 (full precision) or F16 (half, ~50% smaller, slight accuracy loss)
• Memory-Mapped Loading: Large indexes load efficiently without copying into RAM
• CRC32 Validation: Detects corruption on load
• Content Deduplication: Messages are hashed; identical content shares one vector

Index Location: ~/.local/share/coding-agent-search/vector_index/index-<embedder>-<dim>.cvvi

cass supports three search modes, selectable via --mode flag or Alt+S in the TUI:

Lexical Search: Uses Tantivy's BM25 implementation with edge n-grams for prefix matching. Best when you know the exact terms you're looking for.

Semantic Search: Computes vector similarity between query and indexed message embeddings. Finds conceptually related content even without exact term overlap. Requires either the ML model (MiniLM) or falls back to hash embedder.

Hybrid Search: Combines lexical and semantic results using Reciprocal Rank Fusion (RRF):

RRF_score = Σ 1 / (K + rank_i)

Where K=60 (tuning constant) and rank_i is the position in each result list. This balances the precision of lexical search with the recall of semantic search.

# CLI examples
cass search "authentication" --mode lexical --robot
cass search "how to handle user login" --mode semantic --robot
cass search "auth error handling" --mode hybrid --robot

• Wildcard Patterns: Full glob-style pattern support:
  • foo* - Prefix match (finds "foobar", "foo123")
  • *foo - Suffix match (finds "barfoo", "configfoo")
  • *foo* - Substring match (finds "afoob", "configuration")
• Auto-Fuzzy Fallback: When exact searches return sparse results, automatically retries with *term* wildcards to broaden matches. Visual indicator shows when fallback is active.
• Query History Deduplication: Recent searches deduplicated to show unique queries; navigate with Up/Down arrows.
• Match Quality Ranking: New ranking mode (cycle with F12) that prioritizes exact matches over wildcard/fuzzy results.
• Match Highlighting: Use --highlight in robot mode to wrap matching terms with markers (**bold** for text, <mark> for HTML output).

• Three-Pane Layout: Filter bar (top), scrollable results (left), and syntax-highlighted details (right).
• Multi-Line Result Display: Each result shows location and up to 3 lines of context; alternating stripes improve scanability.
• Live Status: Footer shows real-time indexing progress—agent discovery count during scanning, then item progress with sparkline visualization (e.g., 📦 Indexing 150/2000 (7%) ▁▂▄▆█)—plus active filters.
• Multi-Open Queue: Queue multiple results with Ctrl+Enter, then open all in your editor with Ctrl+O. Confirmation prompt for large batches (≥12 items).
• Find-in-Detail: Press / to search within the detail pane; matches highlighted with n/N navigation.
• Mouse Support: Click to select results, scroll panes, or clear filters.
• Theming: Adaptive Dark/Light modes with role-colored messages (User/Assistant/System). Toggle border style (Ctrl+B) between rounded Unicode and plain ASCII.
• Ranking Modes: Cycle through recent/balanced/relevance/quality with F12; quality mode penalizes fuzzy matches.

Ingests history from all major local agents, normalizing them into a unified Conversation -> Message -> Snippet model:

• Codex: ~/.codex/sessions (Rollout JSONL)
• Cline: VS Code global storage (Task directories)
• Gemini CLI: ~/.gemini/tmp (Chat JSON)
• Claude Code: ~/.claude/projects (Session JSONL)
• OpenCode: .opencode directories (SQLite)
• Amp: ~/.local/share/amp & VS Code storage
• Cursor: ~/Library/Application Support/Cursor/User/ global + workspace storage (SQLite state.vscdb)
• ChatGPT: ~/Library/Application Support/com.openai.chat (v1 unencrypted JSON; v2/v3 encrypted—see Environment)
• Aider: ~/.aider.chat.history.md and per-project .aider.chat.history.md files (Markdown)
• Pi-Agent: ~/.pi/agent/sessions (Session JSONL with thinking content)
• Factory (Droid): ~/.factory/sessions (JSONL files organized by workspace slug)

Pi-Agent parses JSONL session files with rich event structure:

• Location: ~/.pi/agent/sessions/ (override with PI_CODING_AGENT_DIR env var)
• Format: Typed events—session_start, message, model_change, thinking_level_change
• Features: Extracts extended thinking content, flattens tool calls with arguments, tracks model changes
• Detection: Scans for *_*.jsonl pattern in sessions directory

OpenCode reads SQLite databases from workspace directories:

• Location: .opencode/ directories (scans recursively from home)
• Format: SQLite database with sessions table
• Detection: Finds directories named .opencode containing database files

Search across agent sessions from multiple machines—your laptop, desktop, and remote servers—all from a single unified index. cass uses SSH/rsync to efficiently sync session data, tracking provenance so you know where each conversation originated.

The easiest way to configure multi-machine search is the interactive setup wizard:

cass sources setup

What the wizard does:

1. Discovers SSH hosts from your ~/.ssh/config
2. Probes each host to check for:
   • Existing cass installation (and version)
   • Agent session data (Claude, Codex, Cursor, Gemini, etc.)
   • System resources (disk space, memory)
3. Lets you select which hosts to configure
4. Installs cass on remotes that don't have it (optional)
5. Indexes existing sessions on remotes (optional)
6. Configures sources.toml with correct paths and mappings
7. Syncs data to your local machine (optional)

Wizard options:

Examples:

# Full interactive wizard
cass sources setup

# Configure specific hosts only
cass sources setup --hosts css,csd,yto

# Preview without making changes
cass sources setup --dry-run

# Resume interrupted setup
cass sources setup --resume

# Non-interactive for CI/CD
cass sources setup --non-interactive --hosts myserver --skip-install

Resumable state: If setup is interrupted (Ctrl+C, connection lost), state is saved to the cache directory (~/.cache/cass/setup_state.json on Linux). Resume with --resume.

When the wizard installs cass on remote machines, it uses an intelligent fallback chain:

Resource Requirements:

• Minimum 2GB disk space
• Recommended 1GB RAM for compilation
• SSH access with key-based authentication

What Gets Installed:

• The cass binary (location depends on method: ~/.cargo/bin/cass for cargo-based, ~/.local/bin/cass for pre-built binary)
• No daemon, no background services—just the binary

Installation Progress: The wizard shows real-time progress for each stage:

Installing cass on laptop...
  [1/4] Checking environment...     ✓
  [2/4] Downloading binary...       ████████░░ 80%
  [3/4] Verifying checksum...       ✓
  [4/4] Setting up PATH...          ✓

Use --skip-install if you prefer to install manually on remotes.

The setup wizard automatically discovers SSH hosts from your configuration:

Discovery Sources:

• ~/.ssh/config (parses Host entries)
• Hosts with wildcards (*, ?) are automatically excluded

Probe Results (for each discovered host):

Probe Caching: Results are cached for 5 minutes to speed up repeated setup attempts. Cache clears automatically on expiry.

For manual configuration without the wizard:

# Add a remote machine using platform presets
cass sources add [email protected] --preset macos-defaults

# Or specify paths explicitly
cass sources add dev@workstation --path ~/.claude/projects --path ~/.codex/sessions

# Sync sessions from all configured sources
cass sources sync

# Check source health and connectivity
cass sources doctor

Sources are configured in the platform config directory (Linux: ~/.config/cass/sources.toml, macOS: ~/Library/Application Support/cass/sources.toml):

[[sources]]
name = "laptop"
type = "ssh"
host = "[email protected]"
paths = ["~/.claude/projects", "~/.codex/sessions"]
sync_schedule = "manual"

[[sources]]
name = "workstation"
type = "ssh"
host = "[email protected]"
paths = ["~/.claude/projects"]
sync_schedule = "daily"

# Path mappings rewrite remote paths to local equivalents
[[sources.path_mappings]]
from = "/home/dev/projects"
to = "/Users/me/projects"

# Agent-specific mappings
[[sources.path_mappings]]
from = "/opt/work"
to = "/Volumes/Work"
agents = ["claude_code"]

Configuration Fields:

# List configured sources
cass sources list [--verbose] [--json]

# Add a new source
cass sources add <user@host> [--name <name>] [--preset macos-defaults|linux-defaults] [--path <path>...] [--no-test]

# Remove a source
cass sources remove <name> [--purge] [-y]

# Check connectivity and config
cass sources doctor [--source <name>] [--json]

# Sync sessions
cass sources sync [--source <name>] [--no-index] [--verbose] [--dry-run] [--json]

The sync engine uses rsync over SSH for efficient delta transfers, with automatic SFTP fallback:

Transfer Methods (auto-detected):

Safety Guarantees:

• Additive-only syncs: rsync runs WITHOUT --delete flag—remote deletions never propagate locally
• No overwrite risk: Existing local files are only updated if remote is newer
• Atomic operations: Failed transfers don't leave partial files

Transfer Configuration:

rsync Flags Used:

-avz --stats --partial --protect-args --timeout=300 \
  -e "ssh -o BatchMode=yes -o ConnectTimeout=10 -o StrictHostKeyChecking=accept-new"

Where -avz = archive mode + verbose + compression.

Data Flow:

Remote: ~/.claude/projects/
    ↓ (rsync over SSH)
Local: ~/.local/share/coding-agent-search/remotes/<source>/<path>/
    ↓ (connector scan)
Index: agent_search.db + tantivy_index/

Where <path> is a filesystem-safe version of the remote path (e.g., .claude_projects).

Sessions from remotes are indexed alongside local sessions, with provenance tracking to identify origin.

When viewing sessions from remote machines, workspace paths may not exist locally. Path mappings rewrite these paths so file links work on your local machine:

# List current mappings
cass sources mappings list laptop

# Add a mapping
cass sources mappings add laptop --from /home/user/projects --to /Users/me/projects

# Test how a path would be rewritten
cass sources mappings test laptop /home/user/projects/myapp/src/main.rs
# Output: /Users/me/projects/myapp/src/main.rs

# Agent-specific mappings (only apply for certain agents)
cass sources mappings add laptop --from /opt/work --to /Volumes/Work --agents claude_code,codex

# Remove a mapping by index
cass sources mappings remove laptop 0

In the TUI, filter sessions by origin:

• F11: Cycle source filter (all → local → remote → all)
• Shift+F11: Open source filter menu to select specific sources

Remote sessions display with a source indicator (e.g., [laptop]) in the results list.

Each conversation tracks its origin:

• source_id: Machine identifier (e.g., "laptop", "workstation")
• source_kind: local or remote
• workspace_original: Original path on the remote machine (before path mapping)

These fields appear in JSON/robot output and enable filtering:

cass search "auth error" --source laptop --json
cass timeline --days 7 --source remote
cass stats --by-source

cass is purpose-built for consumption by AI coding agents—not just as an afterthought, but as a first-class design goal. When you're an AI agent working on a codebase, your own session history and those of other agents become an invaluable knowledge base: solutions to similar problems, context about design decisions, debugging approaches that worked, and institutional memory that would otherwise be lost.

Imagine you're Claude Code working on a React authentication bug. With cass, you can instantly search across:

• Your own previous sessions where you solved similar auth issues
• Codex sessions where someone debugged OAuth flows
• Cursor conversations about token refresh patterns
• Aider chats about security best practices

This cross-pollination of knowledge across different AI agents is transformative. Each agent has different strengths, different context windows, and encounters different problems. cass unifies all this collective intelligence into a single, searchable index.

cass teaches agents how to use it—no external documentation required:

# Quick capability check: what features exist?
cass capabilities --json
# → {"features": ["json_output", "cursor_pagination", "highlight_matches", ...], "connectors": [...], "limits": {...}}

# Full API schema with argument types, defaults, and response shapes
cass introspect --json

# Topic-based help optimized for LLM consumption
cass robot-docs commands # All commands and flags
cass robot-docs schemas # Response JSON schemas
cass robot-docs examples # Copy-paste invocations
cass robot-docs exit-codes # Error handling guide
cass robot-docs guide # Quick-start walkthrough

AI agents sometimes make syntax mistakes. cass aggressively normalizes input to maximize acceptance when intent is clear:

The CLI applies multiple normalization layers:

1. Typo correction: Flags within edit distance 2 are auto-corrected
2. Case normalization: --Robot, --LIMIT → --robot, --limit
3. Single-dash recovery: -robot → --robot (common LLM mistake)
4. Subcommand aliases: find/query/q → search, ls/list → stats
5. Global flag hoisting: Position-independent flag handling

When corrections are applied, cass emits a teaching note to stderr so agents learn the canonical syntax.

Every command supports machine-readable output:

# Pretty-printed JSON (default robot mode)
cass search "error" --robot

# Streaming JSONL: header line with _meta, then one hit per line
cass search "error" --robot-format jsonl

# Compact single-line JSON (minimal bytes)
cass search "error" --robot-format compact

# Include performance metadata
cass search "error" --robot --robot-meta
# → { "hits": [...], "_meta": { "elapsed_ms": 12, "cache_hit": true, "wildcard_fallback": false, ... } }

Design principle: stdout contains only parseable JSON data; all diagnostics, warnings, and progress go to stderr.

LLMs have context limits. cass provides multiple levers to control output size:

Truncated fields include a *_truncated: true indicator so agents know when they're seeing partial content.

Errors are structured, actionable, and include recovery hints:

{
 "error": {
 "code": 3,
 "kind": "index_missing",
 "message": "Search index not found",
 "hint": "Run 'cass index --full' to build the index",
 "retryable": false
 }
}

Exit codes follow a semantic convention:

The retryable field tells agents whether a retry might succeed (e.g., transient I/O) vs. guaranteed failure (e.g., invalid path).

Beyond search, cass provides commands for deep-diving into specific sessions:

# Export full conversation to shareable format
cass export /path/to/session.jsonl --format markdown -o conversation.md
cass export /path/to/session.jsonl --format html -o conversation.html
cass export /path/to/session.jsonl --format json --include-tools

# Expand context around a specific line (from search result)
cass expand /path/to/session.jsonl -n 42 -C 5 --json
# → Shows 5 messages before and after line 42

# Activity timeline: when were agents active?
cass timeline --today --json --group-by hour
cass timeline --since 7d --agent claude --json
# → Grouped activity counts, useful for understanding work patterns

Aggregate search results server-side to get counts and distributions without transferring full result data:

# Count results by agent
cass search "error" --robot --aggregate agent
# → { "aggregations": { "agent": { "buckets": [{"key": "claude_code", "count": 45}, ...] } } }

# Multi-field aggregation
cass search "bug" --robot --aggregate agent,workspace,date

# Combine with filters
cass search "TODO" --agent claude --robot --aggregate workspace

Aggregation Fields:

Response Format:

{
  "aggregations": {
    "agent": {
      "buckets": [
        {"key": "claude_code", "count": 120},
        {"key": "codex", "count": 85}
      ],
      "other_count": 15
    }
  }
}

Top 10 buckets are returned per field, with other_count for remaining items.

Chain multiple searches together by piping session paths from one search to another:

# Find sessions mentioning "auth", then search within those for "token"
cass search "authentication" --robot-format sessions | \
  cass search "refresh token" --sessions-from - --robot

# Build a filtered corpus from today's work
cass search --today --robot-format sessions > today_sessions.txt
cass search "bug fix" --sessions-from today_sessions.txt --robot

How It Works:

1. First search with --robot-format sessions outputs one session path per line
2. Second search with --sessions-from <file> restricts search to those sessions
3. Use - to read from stdin for true piping

Use Cases:

• Drill-down: Broad search → narrow within results
• Cross-reference: Find sessions with term A, then find term B within them
• Corpus building: Save session lists for repeated searches

The --highlight flag wraps matching terms for visual/programmatic identification:

cass search "authentication error" --robot --highlight
# In text output: **authentication** and **error** are bold-wrapped
# In HTML export: <mark>authentication</mark> and <mark>error</mark>

Highlighting is query-aware: quoted phrases like "auth error" highlight as a unit; individual terms highlight separately.

For large result sets, use cursor-based pagination:

# First page
cass search "TODO" --robot --robot-meta --limit 20
# → { "hits": [...], "_meta": { "next_cursor": "eyJ..." } }

# Next page
cass search "TODO" --robot --robot-meta --limit 20 --cursor "eyJ..."

Cursors are opaque tokens encoding the pagination state. They remain valid as long as the index isn't rebuilt.

For debugging and logging, attach a request ID:

cass search "bug" --robot --request-id "req-12345"
# → { "hits": [...], "_meta": { "request_id": "req-12345" } }

For safe retries (e.g., in CI pipelines or flaky networks):

cass index --full --idempotency-key "build-$(date +%Y%m%d)"
# If same key + params were used in last 24h, returns cached result

Debug why a search returned unexpected results:

cass search "auth*" --robot --explain
# → Includes parsed query AST, term expansion, cost estimates

cass search "auth error" --robot --dry-run
# → Validates query syntax without executing

For debugging agent pipelines:

cass search "error" --robot --trace-file /tmp/cass-trace.json
# Appends execution span with timing, exit code, and command details

For machine-readable documentation, use cass robot-docs <topic>:

# Get documentation programmatically
cass robot-docs guide
cass robot-docs schemas
cass robot-docs exit-codes

# Machine-first help (wide output, no TUI assumptions)
cass --robot-help

cass maintains a stable API contract for automation:

cass api-version --json
# → { "version": "0.4.0", "contract_version": "1", "breaking_changes": [] }

cass introspect --json
# → Full schema: all commands, arguments, response types

Contract Version: Currently 1. Increments only on breaking changes.

Guaranteed Stable:

• Exit codes and their meanings
• JSON response structure for --robot output
• Flag names and behaviors
• _meta block format

🔎 cass — Search All Your Agent History

 What: cass indexes conversations from Claude Code, Codex, Cursor, Gemini, Aider, ChatGPT, and more into a unified, searchable index. Before solving a problem from scratch, check if any agent already solved something similar.

 ⚠️ NEVER run bare cass — it launches an interactive TUI. Always use --robot or --json.

 Quick Start

 # Check if index is healthy (exit 0=ok, 1=run index first)
 cass health

 # Search across all agent histories
 cass search "authentication error" --robot --limit 5

 # View a specific result (from search output)
 cass view /path/to/session.jsonl -n 42 --json

 # Expand context around a line
 cass expand /path/to/session.jsonl -n 42 -C 3 --json

 # Learn the full API
 cass capabilities --json # Feature discovery
 cass robot-docs guide # LLM-optimized docs

 Why Use It

 - Cross-agent knowledge: Find solutions from Codex when using Claude, or vice versa
 - Forgiving syntax: Typos and wrong flags are auto-corrected with teaching notes
 - Token-efficient: --fields minimal returns only essential data

 Key Flags

 | Flag | Purpose |
 |------------------|--------------------------------------------------------|
 | --robot / --json | Machine-readable JSON output (required!) |
 | --fields minimal | Reduce payload: source_path, line_number, agent only |
 | --limit N | Cap result count |
 | --agent NAME | Filter to specific agent (claude, codex, cursor, etc.) |
 | --days N | Limit to recent N days |

 stdout = data only, stderr = diagnostics. Exit 0 = success.

cass supports a rich query syntax designed for both humans and machines.

Combine terms with explicit operators for complex queries:

Operator Precedence: NOT binds tightest, then AND, then OR. Use parentheses (in robot mode) for explicit grouping.

# Complex boolean query
cass search "authentication AND (error OR failure) NOT test" --robot

# Exclude test files
cass search "bug fix -test -spec" --robot

# Either error type
cass search "TypeError OR ValueError" --robot

Wrap terms in double quotes for exact phrase matching:

Phrases respect word order and proximity. Useful for error messages, code patterns, and specific terminology.

Tip: Prefix wildcards (foo*) are optimized via pre-computed edge n-grams. Suffix and substring wildcards fall back to regex and are slower on large indexes.

# Field-specific search (in robot mode)
cass search "error" --agent claude --workspace /path/to/project

# Time-bounded search
cass search "bug" --since 2024-01-01 --until 2024-01-31
cass search "bug" --today
cass search "bug" --days 7

# Combined filters
cass search "authentication" --agent codex --workspace myproject --week

cass accepts a wide variety of time/date formats for filtering:

Intelligent Heuristics:

• Numbers >10 digits are treated as milliseconds, otherwise seconds
• Two-digit years are expanded (24 → 2024)
• Date-only inputs default to midnight start or 23:59:59 end

# All equivalent for "last week"
cass search "bug" --since -7d
cass search "bug" --since "-1w"
cass search "bug" --days 7

# Date range
cass search "feature" --since 2024-01-01 --until 2024-01-31

# Mix formats
cass search "error" --since yesterday --until now

Search results include a match_type indicator:

When an exact query returns fewer than 3 results, cass automatically retries with wildcard expansion:

• auth → *auth*
• Results are flagged with wildcard_fallback: true in robot mode
• TUI shows a "fuzzy" indicator in the status bar

The detail pane has three tabs, switchable with [ and ]:

Control how much content shows in the detail preview. Cycle with F7:

Peek Mode (Ctrl+Space): Temporarily expand to XL context. Press again to restore previous size. Useful for quick deep-dives without changing your preferred default.

• Click on result to select
• Click on filter chip to edit/remove
• Scroll in any pane
• Double-click to open result

Efficiently work with multiple search results at once:

Multi-Select Mode:

1. Press m to toggle selection on current result (checkbox appears)
2. Navigate to other results and press m again
3. Press Ctrl+A to select/deselect all visible results
4. Selected count shown in footer: "3 selected"

Bulk Actions Menu (A when items selected):

Multi-Open Queue: For opening many files without navigating away:

1. Press Ctrl+Enter to add current result to queue
2. Continue searching and adding more results
3. Press Ctrl+O to open all queued items
4. Confirmation prompt appears for 12+ items

Clipboard Operations:

• y - Copy current item (cycles: path → snippet → full content)
• Ctrl+Y - Copy all selected items (paths on separate lines)

Cycle through modes with F12:

1. Recent Heavy (default): Strongly favors recent conversations

   • Score = text_relevance × 0.3 + recency × 0.7
   • Best for: "What was I working on?"

2. Balanced: Equal weight to relevance and recency

   • Score = text_relevance × 0.5 + recency × 0.5
   • Best for: General-purpose search

3. Relevance: Prioritizes text match quality

   • Score = text_relevance × 0.8 + recency × 0.2
   • Best for: "Find the best explanation of X"

4. Match Quality: Penalizes fuzzy/wildcard matches

   • Score = text_relevance × 0.7 + recency × 0.2 + match_exactness × 0.1
   • Best for: Precise technical searches

5. Date Newest: Pure chronological order (newest first)

   • Ignores relevance scoring entirely
   • Best for: "Show me all recent activity"

6. Date Oldest: Pure reverse chronological order (oldest first)

   • Ignores relevance scoring entirely
   • Best for: "When did I first work on this?"

• Text Relevance (BM25): Tantivy's implementation of Okapi BM25, considering:

  • Term frequency in document
  • Inverse document frequency across corpus
  • Document length normalization

• Recency: Exponential decay from current time

  • Documents from today: ~1.0
  • Documents from last week: ~0.7
  • Documents from last month: ~0.3

• Match Exactness: Bonus for exact matches vs wildcards

  • Exact phrase: 1.0
  • Prefix match: 0.8
  • Suffix/Substring: 0.5
  • Fuzzy fallback: 0.3

The final score combines all components using mode-specific weights:

Final_Score = BM25_Score × Match_Quality + α × Recency_Factor

Alpha (α) by Ranking Mode:

Match Quality Factors:

Recency Factor: timestamp / max_timestamp normalized to [0, 1].

This formula ensures that "Recent Heavy" mode (default) surfaces your most recent work, while "Relevance Heavy" finds the best explanations regardless of age.

Each connector transforms agent-specific formats into a unified schema:

┌─────────────────┐     ┌──────────────────┐     ┌─────────────────┐
│  Agent Files    │ ──▶ │    Connector     │ ──▶ │  Normalized     │
│  (proprietary)  │     │  (per-agent)     │     │  Conversation   │
└─────────────────┘     └──────────────────┘     └─────────────────┘
     JSONL                   detect()                agent_slug
     SQLite                  scan()                  workspace
     Markdown                                        messages[]
     JSON                                            created_at

Different agents use different role names:

Agents store timestamps inconsistently:

Tool calls, code blocks, and nested structures are flattened for searchability:

// Original (Claude Code)
{"type": "tool_use", "name": "Read", "input": {"path": "/foo/bar.rs"}}

// Flattened for indexing
"[Tool: Read] path=/foo/bar.rs"

The same conversation content can appear multiple times due to:

• Agent file rewrites
• Backup files
• Symlinked directories
• Re-indexing

cass uses a multi-layer deduplication strategy:

1. Message Hash: SHA-256 of (role + content + timestamp)

   • Identical messages in different files are stored once

2. Conversation Fingerprint: Hash of first N message hashes

   • Detects duplicate conversation files

3. Search-Time Dedup: Results are deduplicated by content similarity

   • Even if stored twice, shown once in results

Common low-value content is filtered from results:

• Empty messages
• Pure whitespace
• System prompts (unless searching for them)
• Repeated tool acknowledgments

# Find past solutions for similar errors
cass search "TypeError: Cannot read property" --days 30

# In TUI: F12 to switch to "relevance" mode for best matches

# What has ANY agent said about authentication in this project?
cass search "authentication" --workspace /path/to/project

# Export findings for a new agent's context
cass export /path/to/relevant/session.jsonl --format markdown

# What did I work on today?
cass timeline --today --json | jq '.groups[].conversations'

# TUI: Press Shift+F5 to cycle through time filters

# Find all debugging sessions for a specific file
cass search "debug src/auth/login.rs" --agent claude

# Expand context around a specific line in a session
cass expand /path/to/session.jsonl -n 150 -C 10

# Current agent searches what previous agents learned
cass search "database migration strategy" --robot --fields minimal

# Get full context for a relevant session
cass view /path/to/session.jsonl -n 42 --json

# Export high-quality problem-solving sessions
cass search "bug fix" --robot --limit 100 | \
  jq '.hits[] | select(.score > 0.8)' > training_candidates.json

Press Ctrl+P to open the command palette—a fuzzy-searchable menu of all available actions.

1. Press Ctrl+P to open
2. Type to fuzzy-filter commands
3. Use Up/Down to navigate
4. Press Enter to execute
5. Press Esc to close

Save your current filter configuration to one of 9 slots for instant recall.

• Active filters (agent, workspace, time range)
• Current ranking mode
• The search query

1. Ctrl+P → "Save view to slot N"
2. Ctrl+P → "Load view from slot N"
3. Ctrl+P → "Saved views" to list all slots

Views are stored in tui_state.json and persist across sessions. Clear all saved views with Ctrl+Shift+Del (resets all TUI state).

Control how many lines each search result occupies. Cycle with Shift+D or via the command palette.

The pane automatically adjusts how many results fit based on terminal height and density mode.

cass includes a sophisticated theming system with multiple presets, accessibility-aware color choices, and adaptive styling.

Cycle through 6 built-in theme presets with F2:

All theme colors are validated against WCAG (Web Content Accessibility Guidelines) contrast requirements:

• Text on backgrounds: Minimum 4.5:1 contrast ratio (AA standard)
• Large text/headers: Minimum 3:1 contrast ratio
• Interactive elements: Clear visual distinction from content

The theming engine calculates relative luminance and contrast ratios at runtime to ensure readability across all color combinations.

Conversation messages are color-coded by role for quick visual parsing:

Each agent type (Claude, Codex, Cursor, etc.) also receives a subtle tint, making multi-agent result lists instantly scannable.

Border decorations automatically adapt to terminal width:

Toggle between rounded Unicode and plain ASCII borders with Ctrl+B.

Save important search results with notes and tags for later reference.

• Persistent storage: Bookmarks saved to bookmarks.db (SQLite)
• Notes: Add annotations explaining why you bookmarked something
• Tags: Organize with comma-separated tags (e.g., "rust, important, auth")
• Search: Find bookmarks by title, note, or snippet content
• Export/Import: JSON format for backup and sharing

{
  "id": 1,
  "title": "Auth bug fix discussion",
  "source_path": "/path/to/session.jsonl",
  "line_number": 42,
  "agent": "claude_code",
  "workspace": "/projects/myapp",
  "note": "Good explanation of JWT refresh flow",
  "tags": "auth, jwt, important",
  "snippet": "The token refresh logic should..."
}

Bookmarks are stored separately from the main index:

• Linux: ~/.local/share/coding-agent-search/bookmarks.db
• macOS: ~/Library/Application Support/coding-agent-search/bookmarks.db
• Windows: %APPDATA%\coding-agent-search\bookmarks.db

cass uses a non-intrusive toast notification system for transient feedback—operations complete, errors occur, or state changes without modal dialogs interrupting your workflow.

• Non-Blocking: Toasts appear in a corner without stealing focus
• Auto-Dismiss: Each type has an appropriate display duration
• Message Coalescing: Duplicate messages show a count badge instead of stacking
• Configurable Position: Toasts can appear in any corner (default: top-right)
• Maximum Visible: Limited to 3-5 visible toasts to prevent screen clutter

Toasts feature:

• Color-coded borders: Matches notification type (blue/green/yellow/red)
• Theme-aware: Adapts to current dark/light theme
• Subtle animation: Fade in/out for smooth appearance

To achieve sub-60ms latency on large datasets, cass implements a multi-tier caching strategy in src/search/query.rs:

1. Sharded LRU Cache: The prefix_cache is split into shards (default 256 entries each) to reduce mutex contention during concurrent reads/writes from the async searcher.
2. Bloom Filter Pre-checks: Each cached hit stores a 64-bit Bloom filter mask of its content tokens. When a user types more characters, we check the mask first. If the new token isn't in the mask, we reject the cache entry immediately without a string comparison.
3. Predictive Warming: A background WarmJob thread watches the input. When the user pauses typing, it triggers a lightweight "warm-up" query against the Tantivy reader to pre-load relevant index segments into the OS page cache.

The system is designed for extensibility via the Connector trait (src/connectors/mod.rs). This allows cass to treat disparate log formats as a uniform stream of events.

classDiagram
 class Connector {
 <<interface>>
 +detect() DetectionResult
 +scan(ScanContext) Vec~NormalizedConversation~
 }
 class NormalizedConversation {
 +agent_slug String
 +messages Vec~NormalizedMessage~
 }

 Connector <|-- CodexConnector
 Connector <|-- ClineConnector
 Connector <|-- ClaudeCodeConnector
 Connector <|-- GeminiConnector
 Connector <|-- OpenCodeConnector
 Connector <|-- AmpConnector
 Connector <|-- CursorConnector
 Connector <|-- ChatGptConnector
 Connector <|-- AiderConnector
 Connector <|-- PiAgentConnector
 Connector <|-- FactoryConnector

 CodexConnector ..> NormalizedConversation : emits
 ClineConnector ..> NormalizedConversation : emits
 ClaudeCodeConnector ..> NormalizedConversation : emits
 GeminiConnector ..> NormalizedConversation : emits
 OpenCodeConnector ..> NormalizedConversation : emits
 AmpConnector ..> NormalizedConversation : emits
 CursorConnector ..> NormalizedConversation : emits
 ChatGptConnector ..> NormalizedConversation : emits
 AiderConnector ..> NormalizedConversation : emits
 PiAgentConnector ..> NormalizedConversation : emits
 FactoryConnector ..> NormalizedConversation : emits

• Polymorphic Scanning: The indexer runs connector factories in parallel via rayon, creating fresh Box<dyn Connector> instances that are unaware of each other's underlying file formats (JSONL, SQLite, specialized JSON).
• Resilient Parsing: Connectors handle legacy formats (e.g., integer vs ISO timestamps) and flatten complex tool-use blocks into searchable text.

cass employs a dual-storage strategy to balance data integrity with search performance.

1. Ingestion: Connectors scan proprietary agent files and normalize them into standard structs.
2. Storage (SQLite): The Source of Truth. Data is persisted to a normalized SQLite schema (messages, conversations, agents). This ensures ACID compliance, reliable storage, and supports complex relational queries (stats, grouping).
3. Search Index (Tantivy): The Speed Layer. New messages are incrementally pushed to a Tantivy full-text index. This index is optimized for speed:

• Fields: title, content, agent, workspace, created_at.
• Prefix Fields: title_prefix and content_prefix use Index-Time Edge N-Grams (not stored on disk to save space) for instant prefix matching.
• Deduping: Search results are deduplicated by content hash to remove noise from repeated tool outputs.

flowchart LR
 classDef pastel fill:#f4f2ff,stroke:#c2b5ff,color:#2e2963;
 classDef pastel2 fill:#e6f7ff,stroke:#9bd5f5,color:#0f3a4d;
 classDef pastel3 fill:#e8fff3,stroke:#9fe3c5,color:#0f3d28;
 classDef pastel4 fill:#fff7e6,stroke:#f2c27f,color:#4d350f;
 classDef pastel5 fill:#ffeef2,stroke:#f5b0c2,color:#4d1f2c;

 subgraph Sources["Local Sources"]
 A1[Codex]:::pastel
 A2[Cline]:::pastel
 A3[Gemini]:::pastel
 A4[Claude]:::pastel
 A5[OpenCode]:::pastel
 A6[Amp]:::pastel
 A7[Cursor]:::pastel
 A8[ChatGPT]:::pastel
 A9[Aider]:::pastel
 A10[Pi-Agent]:::pastel
 A11[Factory]:::pastel
 end

 subgraph Remote["Remote Sources"]
 R1["sources.toml"]:::pastel
 R2["SSH/rsync\nSync Engine"]:::pastel2
 R3["remotes/\nSynced Data"]:::pastel3
 end

 subgraph "Ingestion Layer"
 C1["Connectors\nDetect & Scan\nNormalize & Dedupe"]:::pastel2
 end

 subgraph "Dual Storage"
 S1["SQLite (WAL)\nSource of Truth\nRelational Data\nMigrations"]:::pastel3
 T1["Tantivy Index\nSearch Optimized\nEdge N-Grams\nPrefix Cache"]:::pastel4
 end

 subgraph "Presentation"
 U1["TUI (Ratatui)\nAsync Search\nFilter Pills\nDetails"]:::pastel5
 U2["CLI / Robot\nJSON Output\nAutomation"]:::pastel5
 end

 A1 --> C1
 A2 --> C1
 A3 --> C1
 A4 --> C1
 A5 --> C1
 A6 --> C1
 A7 --> C1
 A8 --> C1
 A9 --> C1
 A10 --> C1
 A11 --> C1
 R1 --> R2
 R2 --> R3
 R3 --> C1
 C1 -->|Persist| S1
 C1 -->|Index| T1
 S1 -.->|Rebuild| T1
 T1 -->|Query| U1
 T1 -->|Query| U2

• Non-Blocking: The indexer runs in a background thread. You can search while it works.
• Parallel Discovery: Connector detection and scanning run in parallel across all CPU cores using rayon, significantly reducing startup time when multiple agents are installed.
• Watch Mode: Uses file system watchers (notify) to detect changes in agent logs. When you save a file or an agent replies, cass re-indexes just that conversation and refreshes the search view automatically.
• Real-Time Progress: The TUI footer updates in real-time showing discovered agents during scanning (e.g., "🔍 Discovering (5 agents found)") and indexing progress with sparkline visualization (e.g., "📦 Indexing 150/2000 (7%) ▁▂▄▆█").

The interactive interface (src/ui/tui.rs) is the largest component (~3.5k lines), implementing a sophisticated Immediate Mode architecture using ratatui.

1. Application State: A monolithic struct tracks the entire UI state (search query, cursor position, scroll offsets, active filters, and cached details).
2. Event Loop: A polling loop handles standard inputs (keyboard/mouse) and custom events (Search results ready, Progress updates).
3. Debouncing: User input triggers an async search task via a tokio channel. To prevent UI freezing, we debounce keystrokes (150ms) and run queries on a separate thread, updating the state only when results return.
4. Optimistic Rendering: The UI renders the current state immediately (60 FPS), drawing "stale" results or loading skeletons while waiting for the async searcher.

graph TD
 Input([User Input]) -->|Key/Mouse| EventLoop
 EventLoop -->|Update| State[App State]
 State -->|Render| Terminal
 
 State -->|Query Change| Debounce{Debounce}
 Debounce -->|Fire| SearchTask[Async Search]
 SearchTask -->|Results| Channel
 Channel -->|Poll| EventLoop

Data integrity is paramount. cass treats the SQLite database (src/storage/sqlite.rs) as an append-only log for conversations:

• Immutable History: When an agent adds a message to a conversation, we don't update the existing row. We insert the new message linked to the conversation ID.
• Deduplication: The connector layer uses content hashing to prevent duplicate messages if an agent re-writes a file.
• Versioning: A schema_version meta-table and strict migration path ensure that upgrades (like the recent move to v3) are safe and atomic.

cass is designed to handle index corruption gracefully and recover automatically.

Every Tantivy index stores a schema_hash.json file containing the schema version:

{"schema_hash":"tantivy-schema-v4-edge-ngram-agent-string"}

# Force complete rebuild
cass index --full --force-rebuild

# Check index health
cass health --json

# Diagnostic information
cass diag --verbose

1. Never lose source data: cass only reads agent files, never modifies them
2. Rebuild is always safe: Worst case, re-index from source files
3. Atomic commits: Index writes are transactional
4. Graceful degradation: Search falls back to SQLite if Tantivy fails

cass maintains multiple layers of redundancy to recover from corruption or schema changes:

Schema Hash Versioning: Each Tantivy index stores a schema_hash.json file containing a hash of the current schema definition. On startup:

1. If hash matches → open existing index
2. If hash differs → schema changed, trigger rebuild
3. If file missing/corrupted → assume stale, trigger rebuild

This ensures that version upgrades with schema changes automatically rebuild the index without user intervention.

Automatic Rebuild Triggers:

SQLite as Ground Truth: The SQLite database serves as the authoritative data store. The rebuild_tantivy_from_db() function can reconstruct the entire Tantivy index from SQLite:

// Iterate all conversations from SQLite
// Re-index each message into fresh Tantivy index
// Progress tracked via IndexingProgress for UI feedback

This means users can always recover from Tantivy corruption by running cass doctor --fix or cass index --full --force-rebuild.

The SQLite database uses versioned schema migrations:

Migration Process:

1. On startup, cass checks schema_version in the database
2. If version < current, migrations run automatically
3. Migrations are incremental and non-destructive
4. User data (bookmarks, TUI state, sources.toml) is always preserved

Safe Files (never deleted during rebuild):

• bookmarks.db - Your saved bookmarks
• tui_state.json - UI preferences
• sources.toml - Remote source configuration
• .env - Environment configuration

Backup Policy: Up to 3 backups retained during major rebuilds. Older backups are automatically cleaned.

The --watch flag enables real-time index updates as agent files change.

File change detected
       ↓
[2 second debounce window]  ← Accumulate more changes
       ↓
[5 second max wait]         ← Force flush if changes keep coming
       ↓
Re-index affected files

• Debounce: 2 seconds (wait for burst of changes to settle)
• Max wait: 5 seconds (don't wait forever during continuous activity)

Each file system event is routed to the appropriate connector:

~/.claude/projects/foo.jsonl  → ClaudeCodeConnector
~/.codex/sessions/rollout-*.jsonl → CodexConnector
~/.aider.chat.history.md → AiderConnector

Watch mode maintains watch_state.json:

{
  "last_scan_ts": 1699900000000,
  "watched_paths": [
    "~/.claude/projects",
    "~/.codex/sessions"
  ]
}

• File-level filtering only: When a file is modified, the entire file is re-scanned
• 1-second mtime slack: Accounts for filesystem timestamp granularity
• No per-message filtering: Prevents data loss when new messages are appended

Generate tab-completion scripts for your shell.

Bash:

cass completions bash > ~/.local/share/bash-completion/completions/cass
# Or: cass completions bash >> ~/.bashrc

Zsh:

cass completions zsh > "${fpath[1]}/_cass"
# Or add to ~/.zshrc: eval "$(cass completions zsh)"

Fish:

cass completions fish > ~/.config/fish/completions/cass.fish

PowerShell:

cass completions powershell >> $PROFILE

• Subcommands (search, index, stats, etc.)
• Flags and options (--robot, --agent, --limit)
• File paths for relevant arguments

curl -fsSL https://raw.githubusercontent.com/Dicklesworthstone/coding_agent_session_search/main/install.sh \
  | bash -s -- --easy-mode --verify

cass

On first run, cass performs a full index. You'll see progress in the footer. Search works immediately (falling back to SQLite or partial results until complete).

• Type to search: "python error", "refactor auth", "c++".
• Wildcards: Use foo* (prefix), *foo (suffix), or *foo* (contains) for flexible matching.
• Navigation: Up/Down to select, Right to focus detail pane. Up/Down in search bar navigates query history.
• Filters:
  • F3: Filter by Agent (e.g., "codex").
  • F4: Filter by Workspace/Project.
  • F5/F6: Time filters (Today, Week, etc.).
• Modes:
  • F2: Toggle Dark/Light theme.
  • F12: Cycle ranking mode (recent → balanced → relevance → quality → newest → oldest).
  • Ctrl+B: Toggle rounded/plain borders.
• Actions:
  • Enter: Open original log file in $EDITOR.
  • Ctrl+Enter: Add current result to queue (multi-open).
  • Ctrl+O: Open all queued results in editor.
  • m: Toggle selection on current item.
  • A: Bulk actions menu (when items selected).
  • y: Copy file path or snippet to clipboard.
  • /: Find text within detail pane.
  • Ctrl+Shift+R: Trigger manual re-index (refresh search results).
  • Ctrl+Shift+Del: Reset TUI state (clear history, filters, layout).

Aggregate sessions from your other machines into a unified index:

# Add a remote machine
cass sources add [email protected] --preset macos-defaults

# Sync sessions from all sources
cass sources sync

# Filter by source in TUI: F11 cycles, Shift+F11 opens menu

See Remote Sources (Multi-Machine Search) for full documentation.

The cass binary supports both interactive use and automation.

# Interactive
cass [tui] [--data-dir DIR] [--once]

# Indexing
cass index [--full] [--watch] [--data-dir DIR] [--idempotency-key KEY]

# Search
cass search "query" --robot --limit 5 [--timeout 5000] [--explain] [--dry-run]
cass search "error" --robot --aggregate agent,workspace --fields minimal

# Inspection & Health
cass status --json                    # Quick health snapshot
cass health                           # Minimal pre-flight check (<50ms)
cass capabilities --json              # Feature discovery
cass introspect --json                # Full API schema
cass context /path/to/session --json  # Find related sessions
cass view /path/to/file -n 42 --json  # View source at line

# Session Analysis
cass export /path/to/session --format markdown -o out.md  # Export conversation
cass expand /path/to/session -n 42 -C 5 --json            # Context around line
cass timeline --today --json                               # Activity timeline

# Remote Sources
cass sources add user@host --preset macos-defaults  # Add machine
cass sources sync                                    # Sync sessions
cass sources doctor                                  # Check connectivity
cass sources mappings list laptop                    # View path mappings

# Utilities
cass stats --json
cass completions bash > ~/.bash_completion.d/cass

Commands for troubleshooting, debugging, and understanding system state:

# Health check (fast, <50ms)
cass health --json
# → { "healthy": true, "index_age_seconds": 120, "message_count": 5000 }

# Detailed status with recommendations
cass status --json
# → Includes index freshness, staleness threshold, recommended action

# System diagnostics
cass diag --verbose --json
# → Database stats, index info, connector status, environment

# Query explanation (debug why results are what they are)
cass search "auth" --explain --dry-run --robot
# → Shows parsed query, index strategy, cost estimate without executing

# Find related sessions
cass context /path/to/session.jsonl --json
# → Sessions from same workspace, same day, or same agent

# Comprehensive diagnostic and repair
cass doctor --json
# → Checks data directory, locks, database, index, config files, session sources
# → Reports issues with severity and recommended fixes

# Auto-fix detected issues (safe - only rebuilds derived data)
cass doctor --fix
# → Removes stale locks, rebuilds corrupted index, repairs database

# Force index rebuild even if healthy
cass doctor --fix --force-rebuild
# → Nuclear option: complete index rebuild from source sessions

cass doctor is a comprehensive diagnostic and repair tool designed for troubleshooting installation and data issues. It follows a strict safety philosophy: it never deletes user data.

What it checks:

Safety guarantees:

• Never deletes source agent files - Your Claude, Codex, Cursor sessions are read-only
• Never deletes bookmarks - bookmarks.db is preserved
• Never deletes UI state - tui_state.json is preserved
• Never deletes sources config - sources.toml is preserved
• Only rebuilds derived data - Database and search index can always be regenerated from source sessions

Example output:

$ cass doctor
🩺 cass doctor - Installation Diagnostics

✓ data_directory: OK - Data directory exists and is writable
✓ lock_file: OK - No stale locks found
✓ database: OK - Database accessible (5,234 messages)
✗ index: WARN - Index schema outdated (rebuild recommended)
✓ config: OK - Configuration valid
✓ sources_config: OK - sources.toml parsed successfully
✓ sessions: OK - Found sessions: claude_code, codex, cursor

Summary: 6 passed, 1 warning, 0 failed

💡 Run 'cass doctor --fix' to automatically repair detected issues

Diagnostic Flags:

Commands for managing the semantic search ML model:

# Check current model status
cass models status --json
# → { "installed": true, "model": "all-MiniLM-L6-v2", "size_bytes": 91234567 }

# Install model (downloads ~90MB)
cass models install
# → Downloads from Hugging Face, verifies checksum

# Install from local file (air-gapped environments)
cass models install --from-file /path/to/model.tar.gz

# Verify model integrity
cass models verify --json
# → Checks all required files exist with correct checksums

# Repair corrupted installation
cass models verify --repair
# → Re-downloads missing or corrupted files

# Check for model updates
cass models check-update --json
# → Compares local version with latest available

Model Files (stored in $CASS_DATA_DIR/models/all-MiniLM-L6-v2/):

• model.onnx - The neural network weights (~90MB)
• tokenizer.json - Vocabulary and tokenization rules
• config.json - Model configuration
• special_tokens_map.json - Special token definitions
• tokenizer_config.json - Tokenizer settings

• Verified Install: The installer enforces SHA256 checksums.

• Sandboxed Data: All indexes/DBs live in standard platform data directories (~/.local/share/coding-agent-search on Linux).

• Read-Only Source: cass never modifies your agent log files. It only reads them.

cass uses crash-safe atomic write patterns throughout to prevent data corruption:

TUI State Persistence (tui_state.json):

1. Serialize state to JSON
2. Write to temporary file (tui_state.json.tmp)
3. Atomic rename: temp → final

If a crash occurs during step 2, the original file is untouched. The rename operation (step 3) is atomic on all modern filesystems—it either completes fully or not at all.

ML Model Installation (models/all-MiniLM-L6-v2/):

1. Download to temp directory (models/all-MiniLM-L6-v2.tmp/)
2. Verify all checksums
3. If existing model present: rename to backup (models/all-MiniLM-L6-v2.bak/)
4. Atomic rename: temp → final
5. On success: remove backup
6. On failure: restore from backup

This backup-rename-cleanup pattern ensures that either the old model or new model is always available—never a half-installed state.

Configuration Files (sources.toml, watch_state.json): All configuration writes follow the same temp-file-then-rename pattern, ensuring consistency even during power loss or unexpected termination.

Why This Matters:

• System crashes mid-write won't corrupt your preferences
• Network interruptions during model download won't leave broken installations
• Concurrent processes won't see partially-written files

The project ships with a robust installer (install.sh / install.ps1) designed for CI/CD and local use:

• Checksum Verification: Validates artifacts against a .sha256 file or explicit --checksum flag.

• Rustup Bootstrap: Automatically installs the nightly toolchain if missing.

• Easy Mode: --easy-mode automates installation to ~/.local/bin without prompts.

• Platform Agnostic: Detects OS/Arch (Linux/macOS/Windows, x86_64/arm64) and fetches the correct binary.

cass includes a built-in update checker that notifies you when new versions are available, without interrupting your workflow.

1. Background Check: On TUI startup, a background thread queries GitHub releases
2. Rate Limiting: Checks run at most once per hour to avoid API rate limits
3. Non-Blocking: Update checks never slow down TUI startup or search operations
4. Offline-Safe: Failed network requests are silently ignored

When a new version is available, a notification appears in the TUI with:

• Current version vs. available version
• Release highlights (from GitHub release notes)
• Options: Update Now | Skip This Version | Remind Later

Selecting "Update Now" runs the same verified installer used for initial installation:

# macOS/Linux: Downloads and verifies via curl
curl -fsSL https://...install.sh | bash -s -- --easy-mode --verify

# Windows: PowerShell equivalent
irm https://...install.ps1 | iex

The update process:

1. Downloads the new binary with SHA256 verification
2. Backs up the current binary
3. Replaces with the new version
4. Prompts to restart cass

If you're not ready to update, "Skip This Version" records the skipped version in persistent state. That specific version won't trigger notifications again, but future versions will.

For automated environments or personal preference:

# Environment variable
export CODING_AGENT_SEARCH_NO_UPDATE_PROMPT=1

# Or for fully headless operation
export TUI_HEADLESS=1

Update check state is stored in the data directory:

• last_update_check: Timestamp of last check (for rate limiting)
• skipped_version: Version user chose to skip
• Both are reset on manual update or by deleting tui_state.json

• Config: Loads .env via dotenvy::dotenv().ok(); configure API/base paths there. Do not overwrite .env.

• Data Location: Defaults to standard platform data directories (e.g., ~/.local/share/coding-agent-search). Override with CASS_DATA_DIR or --data-dir.

• ChatGPT Support: The ChatGPT macOS app stores conversations in versioned formats:

  • v1 (legacy): Unencrypted JSON in conversations-{uuid}/ — fully indexed.
  • v2/v3: Encrypted with AES-256-GCM, key stored in macOS Keychain (OpenAI-signed apps only) — detected but skipped.

  Encrypted conversations require keychain access which isn't available to third-party apps. Legacy unencrypted conversations are indexed automatically.

• Logs: Written to cass.log (daily rotating) in the data directory.

• Updates: Interactive TUI checks for GitHub releases on startup. Skip with CODING_AGENT_SEARCH_NO_UPDATE_PROMPT=1 or TUI_HEADLESS=1.

• Cache tuning: CASS_CACHE_SHARD_CAP (per-shard entries, default 256) and CASS_CACHE_TOTAL_CAP (total cached hits across shards, default 2048) control prefix cache size; raise cautiously to avoid memory bloat.

• Cache debug: set CASS_DEBUG_CACHE_METRICS=1 to emit cache hit/miss/shortfall/reload stats via tracing (debug level).

• Watch testing (dev only): cass index --watch --watch-once path1,path2 triggers a single reindex without filesystem notify (also respects CASS_TEST_WATCH_PATHS for backward compatibility); useful for deterministic tests/smoke runs.

• Checksum mismatch: Ensure .sha256 is reachable or pass --checksum explicitly. Check proxies/firewalls.

• Binary not on PATH: Append ~/.local/bin (or your --dest) to PATH; re-open shell.

• Nightly missing in CI: Set RUSTUP_INIT_SKIP=1 if toolchain is preinstalled; otherwise allow installer to run rustup.

• Watch mode not triggering: Confirm watch_state.json updates and that connector roots are accessible; notify relies on OS file events (inotify/FSEvents).

• Reset TUI state: Run cass tui --reset-state (or press Ctrl+Shift+Del in the TUI) to delete tui_state.json and restore defaults.

We target Rust Nightly to leverage the latest optimizations.

# Format & Lint
cargo fmt --check
cargo clippy --all-targets -- -D warnings

# Build & Test
cargo build --release
cargo test

# Run End-to-End Tests
cargo test --test e2e_index_tui
cargo test --test install_scripts

The release profile is aggressively optimized for binary size and performance:

[profile.release]
lto = true              # Link-time optimization across all crates
codegen-units = 1       # Single codegen unit for better optimization
strip = true            # Remove debug symbols from binary
panic = "abort"         # Smaller panic handling (no unwinding)
opt-level = "z"         # Optimize for size over speed

Trade-offs:

• Build time is significantly longer (~3-5x)
• Binary size is ~40-50% smaller
• No stack traces on panic (use debug builds for development)

The CI pipeline (.github/workflows/ci.yml) runs on every PR and push to main:

Coverage Reports:

• lcov.info - LCOV format for tools like codecov
• coverage-summary.txt - Human-readable summary
• Coverage % shown in GitHub Actions step summary

Test Artifacts:

• Trace files from --trace-file runs
• Test run summary logs
• Retained for 7 days (e2e) / 30 days (coverage)

# Generate coverage locally
cargo install cargo-llvm-cov
cargo llvm-cov --all-features --workspace --text

# Run specific e2e tests
cargo test --test e2e_filters -- --test-threads=1

The TUI automatically saves your preferences to tui_state.json in the data directory.

• Linux: ~/.local/share/coding-agent-search/tui_state.json
• macOS: ~/Library/Application Support/coding-agent-search/tui_state.json
• Windows: %APPDATA%\coding-agent-search\tui_state.json

State is saved using atomic file operations: data is first written to a temporary file (tui_state.json.tmp), then atomically renamed to the final location. This prevents corruption if the application crashes mid-save or during unexpected termination. See Atomic File Operations for technical details.

# Via CLI flag
cass tui --reset-state

# Via keyboard (in TUI)
Ctrl+Shift+Del

This deletes tui_state.json and restores all defaults.

• Schema v4 (hash tantivy-schema-v4-edge-ngram-preview) stores agent/workspace/source_path/msg_idx/created_at/title/content plus edge-ngrams (title_prefix, content_prefix) for type-ahead matching.
• New preview field keeps a short, stored excerpt (~200 chars + ellipsis) so prefix-only queries can render snippets without pulling full content.
• Rebuilds auto-trigger when the schema hash changes; index directory is recreated as needed. Tokenizer: hyphen_normalize to keep “cma-es” searchable while enabling prefix splits.

• Wildcard patterns: WildcardPattern enum supports Exact, Prefix (foo*), Suffix (*foo), and Substring (foo). Prefix uses edge n-grams; suffix/substring use Tantivy RegexQuery with escaped special characters.
• Auto-fuzzy fallback: search_with_fallback() wraps the base search; if results < threshold and query has no wildcards, retries with *term* patterns and sets wildcard_fallback flag for UI indicator.
• Cache-first: per-agent + global LRU shards (env CASS_CACHE_SHARD_CAP, default 256). Cached hits store lowered content/title/snippet and a 64-bit bloom mask; bloom + substring keeps validation fast.
• Fallback order: Tantivy (primary) → SQLite FTS (consistency) with deduping/noise filtering. Prefix-only snippet path tries cached prefix snippet, then a cheap local snippet, else Tantivy SnippetGenerator.
• Warm worker: runtime-aware, debounced (env CASS_WARM_DEBOUNCE_MS, default 120 ms), runs a tiny 1-doc search to keep the reader hot; reloads are debounced (300 ms) and counted in metrics (cache hit/miss/shortfall/reloads tracked internally).

• Opens SQLite + Tantivy; --full clears tables/FTS and wipes Tantivy docs; --force-rebuild recreates index dir when schema changes.
• Parallel connector loop: detect → scan runs concurrently across all connectors using rayon's parallel iterator, with atomic progress counters updating discovered agent count and conversation totals in real-time. Ingestion into SQLite and Tantivy happens sequentially after all scans complete. Watch mode: debounced filesystem watcher, path classification per connector, since_ts tracked in watch_state.json, incremental reindex of touched sources. TUI startup spawns a background indexer with watch enabled.

The TUI footer displays live indexing progress through the IndexingProgress struct:

Phase Indicators:

Progress Components:

• Agent Discovery: During scanning, shows which agents were found (e.g., "Found: claude_code, codex, cursor")
• Conversation Count: Total conversations discovered across all agents
• Sparkline Visualization: Mini bar chart showing indexing velocity (▁▂▄▆█)
• Rebuild Indicator: Special badge when performing full schema rebuild vs. incremental update

Implementation Details:

pub struct IndexingProgress {
    pub total: AtomicUsize,              // Total items to process
    pub current: AtomicUsize,            // Items processed so far
    pub phase: AtomicUsize,              // 0=Idle, 1=Scanning, 2=Indexing
    pub is_rebuilding: AtomicBool,       // Full rebuild vs incremental
    pub discovered_agents: AtomicUsize,  // Count of agents found
    pub discovered_agent_names: Mutex<Vec<String>>,  // Agent names for display
    pub last_error: Mutex<Option<String>>,  // Background indexer errors
}

All counters use AtomicUsize with relaxed ordering for lock-free updates from the background indexer thread. The TUI polls these at ~10Hz for smooth progress display without synchronization overhead.

• Normalized relational model (agents, workspaces, conversations, messages, snippets, tags) with FTS mirror on messages. Single-transaction insert/upsert, append-only unless --full. schema_version guard; bundled modern SQLite.

• Three-pane layout (agents → results → detail), responsive splits, focus model (Tab/Shift+Tab), mouse support. Detail tabs (Messages/Snippets/Raw) plus full-screen modal with role colors, code blocks, JSON pretty-print, highlights. Footer packs shortcuts + mode badges; state persisted in tui_state.json.

• Each connector implements detect (root discovery) and scan (since_ts-aware ingestion). External IDs preserved for dedupe; workspace/source paths carried through; roles normalized.

• Checksum-verified easy/normal modes, optional quickstart (index on first run), rustup bootstrap if needed. PATH hints appended with warnings; SHA256 required.

• Benches: index_perf measures full index build; runtime_perf covers search latency + indexing micro-cases.
• Tests: unit + integration + headless TUI e2e; installer checksum fixtures; watch-mode and index/search integration; cache/bloom UTF-8 safety and bloom gate tests.

1. Speed Over Space

cass makes deliberate trade-offs favoring query latency over storage efficiency:

• Edge N-grams: We pre-compute prefix substrings (length 2-20) for every token during indexing. This multiplies index size but enables O(1) prefix lookups instead of O(n) regex scans.
• Dual Storage: Data lives in both SQLite (relational queries, ACID guarantees) and Tantivy (full-text search). This redundancy costs disk space but provides optimal performance for each access pattern.
• Bloom Filter Caching: Each cached search hit stores a 64-bit bloom mask plus lowercase copies of content/title/snippet. Memory cost per cached hit: ~500 bytes. Benefit: sub-millisecond cache filtering.

2. Local-First, Privacy by Design

Your coding sessions contain sensitive information—API keys, internal codenames, debugging strategies. cass is architected to never transmit data:

• No telemetry, analytics, or crash reporting
• No network calls except optional GitHub release checks (easily disabled)
• Index and database stored in user-controlled directories
• Source agent files are read-only—never modified

3. Resilience Over Strictness

AI agents make mistakes. Humans make typos. cass is aggressively forgiving:

• Typo Correction: Levenshtein distance ≤2 flags are auto-corrected with teaching feedback
• Case Normalization: --Robot, --LIMIT → --robot, --limit
• Command Aliases: find/query/q all resolve to search
• Graceful Degradation: Encrypted ChatGPT conversations are detected and skipped rather than crashing

4. Incremental by Default

Full reindexing is expensive. cass minimizes work through careful state tracking:

• File Modification Times: Connectors skip unchanged files using mtime comparison with 1-second slack for filesystem granularity
• Append-Only Messages: When a conversation grows, only new messages (where idx > max_existing_idx) are inserted
• Watch State Persistence: Per-connector timestamps in watch_state.json enable surgical re-scanning

Traditional full-text search requires expensive wildcard expansion for prefix queries. cass inverts this by pre-computing all prefixes at index time:

Input token: "authentication"
Generated n-grams: "au", "aut", "auth", "authe", "authen", "authent", "authenti",
                   "authentic", "authentica", "authenticat", "authenticati",
                   "authenticatio", "authentication"

Implementation (src/search/tantivy.rs:288-305):

• N-gram lengths: 2 to 20 characters
• Stored in separate Tantivy fields: title_prefix, content_prefix
• These fields are indexed but NOT stored (saves disk space)
• Query "auth*" becomes a simple term lookup on the prefix field

Performance Impact:

• Index build time: ~20% slower
• Index size: ~3x larger
• Query time for prefixes: O(1) instead of O(n) regex scan

When filtering cached search results against a refined query, string comparison is expensive. cass uses a 64-bit bloom filter as a fast negative gate:

fn hash_token(tok: &str) -> u64 {
    let mut h: u64 = 5381;  // djb2 initial value
    for b in tok.as_bytes() {
        h = ((h << 5).wrapping_add(h)).wrapping_add(u64::from(*b));
    }
    1u64 << (h % 64)  // Map to single bit position
}

How It Works:

1. During caching, compute bloom64 = hash(token1) | hash(token2) | ... for all content tokens
2. On cache lookup, compute query bloom mask the same way
3. If (cached.bloom64 & query_bloom) != query_bloom, the cached hit cannot match—skip expensive string comparison
4. If bloom passes, proceed with actual substring matching

False Positive Rate: With 64 bits and typical 5-15 tokens per query, ~70% of non-matching hits are rejected by bloom alone.

Tantivy provides BM25 (Best Match 25) scoring out of the box. cass extends this with:

Match Type Quality Factors:

Ranking Mode Formulas:

Recent Heavy:    score = bm25 × 0.3 + recency × 0.7
Balanced:        score = bm25 × 0.5 + recency × 0.5
Relevance Heavy: score = bm25 × 0.8 + recency × 0.2
Match Quality:   score = bm25 × 0.7 + recency × 0.2 + quality_factor × 0.1

Where recency is an exponential decay: e^(-age_days / decay_constant)

cass uses Rayon's work-stealing thread pool for parallel agent discovery:

let pending_batches: Vec<_> = connector_factories
    .into_par_iter()
    .filter_map(|(name, factory)| {
        let conn = factory();  // Each thread gets fresh instance
        if !conn.detect().detected { return None; }
        conn.scan(&ctx).ok().map(|convs| (name, convs))
    })
    .collect();  // Parallel collection

Why This Matters:

• 11 connectors × ~100ms average scan time = 1100ms sequential
• With 4 cores: ~250ms parallel (3.6x speedup)
• Atomic counters provide lock-free progress updates to UI

cass automatically selects the optimal query strategy based on pattern:

Auto-Fallback Logic: When exact search returns <3 results and query has 1-2 terms, automatically retry with *term* wildcards. Results are flagged with wildcard_fallback: true.

Total typical footprint: 70-140MB for a 50K message corpus.

cass is designed for individual developer use (1K-500K messages). Beyond that:

What cass Reads:

• Agent session files (JSONL, JSON, SQLite, Markdown)
• File modification times (for incremental indexing)
• Environment variables (configuration only)

What cass Writes:

• ~/.local/share/coding-agent-search/ (or platform equivalent):
  • agent_search.db - SQLite database
  • tantivy_index/ - Full-text search index
  • models/all-MiniLM-L6-v2/ - Optional local semantic model files
  • vector_index/ - Optional semantic vector index (.cvvi)
  • tui_state.json - UI preferences
  • watch_state.json - Incremental index state
  • cass.log - Rotating log file

What cass NEVER Does:

• Modify source agent files (strictly read-only)
• Make network requests (except optional update checks or explicit user-initiated model installs)
• Execute code from indexed content
• Access files outside known agent directories

ChatGPT Encrypted Conversations:

• Versions 2 and 3 of the ChatGPT macOS app use AES-256-GCM encryption
• Keys are stored in the macOS Keychain, accessible only to OpenAI-signed apps
• cass detects encrypted files and gracefully skips them
• Optional: Provide your own key via CHATGPT_ENCRYPTION_KEY (base64) or ~/.config/cass/chatgpt_key.bin

No Sensitive Data in Logs:

• Log files contain operation traces, not message content
• Error messages are sanitized to avoid leaking paths/content

cass assumes:

• The local filesystem is trusted
• The user running cass should have access to all indexed content
• Network is untrusted (hence no network calls)

cass does NOT protect against:

• Root/admin access to the machine
• Memory forensics while running
• Physical access to the storage device

Verdict: grep is better for one-off searches in known files. cass excels at exploring your entire coding history across agents.

Verdict: Cloud search makes sense for team collaboration. cass is for individual developers who want speed and privacy.

cass actually uses SQLite FTS5 as a fallback! But Tantivy provides:

Verdict: SQLite FTS5 is great for simple search. Tantivy handles the sophisticated queries cass needs.

Every agent connector implements this interface (src/connectors/mod.rs):

pub trait Connector {
    /// Check if this agent's data exists on the system
    fn detect(&self) -> DetectionResult;

    /// Scan and normalize all conversations (respecting since_ts for incremental)
    fn scan(&self, ctx: &ScanContext) -> anyhow::Result<Vec<NormalizedConversation>>;
}

pub struct DetectionResult {
    pub detected: bool,
    pub root_paths: Vec<PathBuf>,  // Where to watch for changes
    pub version: Option<String>,
}

pub struct ScanContext {
    pub data_dir: PathBuf,         // Primary data directory (cass internal state)
    pub scan_roots: Vec<ScanRoot>, // Scan roots for agent logs (empty = use default detection)
    pub since_ts: Option<i64>,     // Only scan files modified after this timestamp
}

Step 1: Create the connector file (src/connectors/my_agent.rs):

use super::*;

pub struct MyAgentConnector;

impl Connector for MyAgentConnector {
    fn detect(&self) -> DetectionResult {
        let root = dirs::home_dir()
            .map(|h| h.join(".my-agent/sessions"));

        DetectionResult {
            detected: root.as_ref().map(|p| p.exists()).unwrap_or(false),
            root_paths: root.into_iter().collect(),
            version: None,
        }
    }

    fn scan(&self, ctx: &ScanContext) -> anyhow::Result<Vec<NormalizedConversation>> {
        // Determine scan root - check if data_dir looks like our storage (for testing)
        // or fall back to default detection
        let looks_like_storage = |p: &PathBuf| {
            p.to_str().map(|s| s.contains("my-agent")).unwrap_or(false)
        };

        let root = if ctx.use_default_detection() {
            if looks_like_storage(&ctx.data_dir) && ctx.data_dir.exists() {
                ctx.data_dir.clone()
            } else {
                // Fall back to default location
                match dirs::home_dir().map(|h| h.join(".my-agent/sessions")) {
                    Some(r) if r.exists() => r,
                    _ => return Ok(Vec::new()),
                }
            }
        } else {
            // Explicit scan_roots provided
            ctx.scan_roots.first()
                .map(|sr| sr.path.clone())
                .unwrap_or_else(|| ctx.data_dir.clone())
        };

        let mut conversations = Vec::new();

        // Find session files
        for entry in walkdir::WalkDir::new(&root) {
            let entry = entry?;
            if !entry.path().extension().map(|e| e == "json").unwrap_or(false) {
                continue;
            }

            // Skip unchanged files for incremental indexing
            if let Some(since) = ctx.since_ts {
                if !file_modified_since(entry.path(), since) {
                    continue;
                }
            }

            // Parse and normalize
            let conv = parse_my_agent_file(entry.path())?;
            conversations.push(conv);
        }

        Ok(conversations)
    }
}

Step 2: Register in the connector factory (src/indexer/mod.rs):

let connector_factories: Vec<(&'static str, fn() -> Box<dyn Connector + Send>)> = vec![
    ("codex", || Box::new(CodexConnector::new())),
    ("my_agent", || Box::new(MyAgentConnector)),  // Add here
    // ...
];

Step 3: Add watch path classification (src/indexer/mod.rs:classify_paths):

if path_str.contains(".my-agent") {
    kinds.insert(ConnectorKind::MyAgent);
}

When converting agent-specific formats to NormalizedConversation:

1. Roles: Map to "user", "assistant", "system", "tool", or "agent"
2. Timestamps: Convert to milliseconds since Unix epoch (i64)
3. External ID: Use a stable identifier unique within the agent (file path, UUID, etc.)
4. Content Flattening: Extract searchable text from nested structures:

   // Tool calls become searchable text
   "[Tool: ReadFile] path=/src/main.rs, lines=1-50"

5. Workspace Detection: Extract working directory from session metadata when available

• Semantic Search: Embed conversations using local models (Ollama integration) for meaning-based retrieval
• Session Grouping: Automatically cluster related conversations by project/topic
• Export Improvements: Better markdown/HTML export with syntax highlighting
• Windows Native: First-class Windows support with native installer

• MCP Server Mode: Run as a Model Context Protocol server for direct agent integration
• Conversation Analytics: Token usage tracking, agent comparison dashboards
• Collaborative Features: Optional encrypted sync between machines
• Plugin System: User-defined connectors without recompiling

• Agent Memory Layer: Let AI agents query their own history as working memory
• Team Knowledge Base: Shared, searchable repository of team coding sessions
• Learning from History: Surface patterns in how problems were solved across sessions

These are explicitly out of scope:

• Cloud hosting: cass will always be local-first
• Real-time collaboration: That's a different tool category
• Agent execution: cass searches history, it doesn't run agents
• Universal file search: Focus remains on AI coding agent sessions

MIT. See LICENSE for details.