Audrey
Audrey is a local-first persistent memory layer for AI agents (including Claude Code). It encodes episodic events into a SQLite-backed store (with vector search via sqlite-vec), recalls relevant memories per prompt, and runs maintenance (“dream”/consolidation) to consolidate episodes into semantic/procedural knowledge with confidence decay and contradiction/validation handling. It can be used as a JS SDK, an MCP server with Claude Code memory tools, a CLI with lifecycle hooks, and a local REST API server.
Score Breakdown
⚙ Agent Friendliness
🔒 Security
REST server supports Bearer token auth via AUDREY_API_KEY, but the README does not document TLS requirements, authorization model, or fine-grained permissions. Secrets are referenced via environment variables for LLM/embeddings (good practice implied), but there’s no stated guidance on avoiding secret leakage in logs. Local-first storage reduces external data exposure, yet consolidation/embeddings may transmit data to external providers if configured.
⚡ Reliability
Best When
You want local, reviewable agent memory on a single machine (or per-user deployment) with MCP/CLI hooks for Claude Code, plus optional embeddings/LLM providers for consolidation and maintenance.
Avoid When
You need centralized managed memory with enterprise-grade tenancy controls, or you cannot accept that embeddings/LLM consolidation may call external providers (when configured).
Use Cases
- • Local, durable memory for multi-session AI agents (especially Claude Code)
- • Per-prompt semantic memory injection to improve context continuity
- • Consolidating session learnings into reusable principles/workflows (dream/consolidation)
- • Detecting/handling contradictory facts and tracking resolution state
- • Building agent applications needing reviewable, exportable memory snapshots (JSON)
- • Standalone memory backend for any framework via REST API
Not For
- • A compliance boundary (must be paired with app-level access control and audit/retention controls)
- • High-availability, multi-region production storage without additional infrastructure (it’s local-first)
- • Use cases requiring strict external data residency controls without hosting guarantees
- • Teams needing a strongly standardized, published OpenAPI/typed contract beyond what the CLI/README describes
Interface
Authentication
Auth is mentioned for the REST server using a single Bearer token environment variable. The README does not describe fine-grained scopes/permissions for tools.
Pricing
No hosted service pricing is described; package is local-first.
Agent Metadata
Known Gotchas
- ⚠ Memory is stored locally in a SQLite file; concurrent agent/worker access may require careful process coordination (not documented here).
- ⚠ REST server appears local (default port 3487); exposing to networks without strong deployment security may be risky.
- ⚠ When using hosted embeddings/LLMs for consolidation, failures/timeouts and provider rate limits should be handled by the integrator (no explicit retry/rate-limit guidance described in README).
- ⚠ Because embeddings are re-generated on snapshot restore, cost/latency can spike depending on embedding provider configuration.
Alternatives
Full Evaluation Report
Comprehensive deep-dive: security analysis, reliability audit, agent experience review, cost modeling, competitive positioning, and improvement roadmap for Audrey.
AI-powered analysis · PDF + markdown · Delivered within 30 minutes
Package Brief
Quick verdict, integration guide, cost projections, gotchas with workarounds, and alternatives comparison.
Delivered within 10 minutes
Score Monitoring
Get alerted when this package's AF, security, or reliability scores change significantly. Stay ahead of regressions.
Continuous monitoring
Scores are editorial opinions as of 2026-03-30.