betterdb-agent-cache v0.11.0

Adds the Pydantic AI adapter (betterdb_agent_cache.adapters.pydantic_ai) —
exact-match LLM caching for Pydantic AI agents on plain Valkey.

- CachedModel(model, cache): wraps any Pydantic AI Model, intercepts request()
  for cache-before-call and delegates everything else via __getattr__. On a hit
  the response (text or tool calls) is reconstructed with stored token counts
  and the underlying model is not called.
- prepare_params(): normalizes Pydantic AI message history to LlmCacheParams
  across all part types. ThinkingPart is dropped as non-deterministic; ImageUrl
  and BinaryContent are routed through the binary normalizer so blobs become
  compact refs instead of raw base64.
- model_request_parameters (tool schemas) is excluded from the cache key —
  safe when one CachedModel wraps a single Agent.
- New optional extra: pydantic_ai (pydantic-ai-slim>=2.0.0).

Contributed by the community (#131).

@betterdb/semantic-cache v0.10.0 — add Google AI (Gemini) embedding p…

…rovider (#160)

betterdb-semantic-cache v0.8.0 — add Google AI (Gemini) embedding pro…

…vider (#160)

@betterdb/mcp v1.3.1

Fixes:
- report package version from metadata (#222)
- pass section param through to info endpoint (#280)

@betterdb/agent-memory v0.6.0

Opt-in write-time fact consolidation (#299): MemoryStore.consolidateFacts()
distills memories into atomic, deduplicated, dated facts (additive, non-lossy),
with subject-keyed reconciliation. Off by default behind a constructor flag.

betterdb-agent-memory (Python) v0.5.0

Python parity port of opt-in write-time fact consolidation (#299):
MemoryStore.consolidate_facts() + reconcile_facts. Off by default.

@betterdb/agent-cache v0.11.1

Build fix: exclude nested test dirs (src/**/__tests__) so src/adapters/__tests__
no longer compiles into the published dist/. No runtime or API changes.

betterdb-agent-cache (Python) v0.10.0

Added: OpenAI Agents SDK adapter (#149) —
betterdb_agent_cache.adapters.openai_agents (CachedModel / CachedModelProvider),
caching at the Model.get_response() level. New optional extra: openai_agents.

cleanup

chore(agent-cache): release v0.11.0 (streaming cache) + fix TS build (#…

…306)

#262 shipped wrapStream (streaming response caching for the Vercel AI SDK
adapter) but left the package version at 0.10.0, filed the changelog entry
under a stale "[0.8.0] - Unreleased" heading, and didn't typecheck cleanly.

Release:
- Bump packages/agent-cache to 0.11.0 (new backward-compatible feature).
- Retitle the changelog entry to [0.11.0] - 2026-07-07.

TS build fixes (package tsc was failing):
- Add @ai-sdk/provider as a direct devDependency — ai.ts imports its
  LanguageModelV3* types, which pnpm's strict layout hid as a transitive dep.
- ai.test.ts: type the mock params as LanguageModelV3CallOptions and add a
  streamCall() helper that supplies the doGenerate/model fields the middleware
  call signature requires (tests exercise only the streaming path).

Docs:
- The Vercel AI SDK adapter now caches doStream in addition to doGenerate;
  move streaming out of "Known limitations" (only tool-call streams remain
  uncached, and cached streams replay as a single chunk).