3 releases
| new 0.1.4 | Jan 9, 2026 |
|---|---|
| 0.1.3 | Jan 9, 2026 |
| 0.1.2 | Jan 9, 2026 |
#13 in #llm-agent
230KB
6K
SLoC
OsmoGrep
A terminal-native execution agent that validates code changes by running real tests, not eyeballing diffs.
Can your code changes be validated autonomously through safe, deterministic execution?
Functionality?
| Area | What OsmoGrep Does |
|---|---|
| Input | Uncommitted git diffs |
| Context | AST-based symbol extraction + test graph |
| Action | Generates + runs real tests |
| Isolation | Executes in temporary agent branches only |
| Validation | Single-test or full-suite execution |
| Artifacts | Logs, reports, test files, structured output |
| UI | Highly optimized terminal TUI |
Everything runs in an isolated agent branch, fully reversible and safe.
Supports:
- Local models (Ollama)
- OpenAI-compatible APIs
Core Capabilities
| Capability | Description |
|---|---|
| Diff Inspector | Parses diffs, extracts symbols, metrics, risk |
| Context Graph | AST-driven symbol and call-graph extraction |
| Single-Test Generation | LLM generates minimal reproduction tests |
| Full Suite Execution | Runs repo tests and builds a structured report |
| Deterministic Cache | Avoids regenerating tests that already passed |
| Sandboxed Execution | Uses OS-level worktrees for isolation |
| Artifacts | Test files, logs, markdown, machine-readable JSON |
Designed for real engineering workflows—not toy demos.
Installation
curl -fsSL https://raw.githubusercontent.com/kaushal07wick/OsmoGrep/master/install.sh | sh
Then run inside any git repository:
osmogrep
Usage
Write code → git add . → run:
osmogrep
The agent inspects your diff, builds context, and executes in a sandbox branch.
Commands Overview
| Command | Description |
|---|---|
help |
Show all commands |
inspect |
Analyze git changes and build context |
changes |
List analyzed changes |
changes <n> |
Open diff in split TUI |
agent run <n> |
Generate + run test for specific change |
agent run --all |
Generate tests for all diffs (parallel mode) |
agent run --full |
Execute entire test suite with report |
model use openai/anthropic <model> |
Configure remote model |
model use ollama <model> |
Use local Ollama model |
model show |
Show active model |
agent cancel |
Cancel running agent |
agent status |
Show current status |
branch list |
List available git branches |
clear / logs clear |
Clear logs |
close |
Close result panel |
quit |
Exit OsmoGrep |
License
Licensed under MIT. See the LICENSE file for details.
Dependencies
~26–44MB
~775K SLoC