MemoryAtlas

Dual-mode memory infrastructure: Local mode stores an atomic disk snapshot plus append-only journal under ~/.retaindb with no external dependencies (no Postgres, Redis, or Qdrant). Server/Cloud mode uses PostgreSQL with pgvector. Both modes share the same retrieval pipeline: BM25 lexical search + vector similarity + knowledge-graph signals (updates/contradicts/supports/extends/derives relations), fused with RRF then reranked. Supports 12 typed memory categories with temporal validity (validFrom/validUntil), recall reinforcement (access counts + memory-strength decay), and token-budgeted delta compression for context packing.

Storage
Local mode: atomic disk snapshot + append-only journal under ~/.retaindb (zero external dependencies; optional local-transformers or hash-vector fallback for embeddings). Server/Cloud mode: PostgreSQL with pgvector extension; embeddings via local-transformers, OpenAI, or hash-vector fallback.
Retrieval
BM25 lexical search + vector similarity + knowledge-graph signals fused with RRF, then reranked. Context-pack endpoint builds token-budgeted responses with delta compression — sends only chunks changed since a prior pack hash, tracked via context_hash/delta_context. MCP tools expose context, recall, remember, handoff, session_history, and forget operations. Graph relations (updates, contradicts, supports, extends, derives) add temporal supersession and cross-memory reasoning.
Self-host
Self-host: moderate
License
Apache-2.0 / BSL-1.1
Pricing
Free tier: 20k writes + 5k context reads/mo (no credit card); Builder $9/mo (75k writes, Knowledge Graph, delta compression); Pro $29/mo (250k writes, 10 projects); Scale $99/mo (1M writes, unlimited projects); enterprise custom. Local/self-hosted server is free. · Freemium
GitHub stars
46
Last release
Last commit
2026-06-12
First catalogued
2026-06-28

Strengths

  • Dual local/cloud deployment: Local mode is a single npx command (npx -y @retaindb/local) with no cloud account; Server/Cloud mode adds PostgreSQL + pgvector, multi-user auth, and external knowledge-source connectors
  • 12 typed memory categories (factual, preference, procedural, decision, correction, constraint, instruction, goal, event, session_summary, project_state, semantic) with temporal validity (validFrom/validUntil) and graph relations for supersession and contradiction tracking
  • Token-efficient context packing: token-budgeted context packs with delta compression send only chunks changed since the prior pack hash, cutting repeated-context overhead for long-running coding agents
  • Multi-modal retrieval (BM25 + vector + graph signals + RRF + rerank) catches exact-token matches — codes, IDs, error strings — that pure semantic search misses
  • Named native support for coding-agent clients (Codex, Claude Code, OpenCode) via retaindb connect all and MCP; session replay viewer and concept-graph UI bundled with the local runtime

Watch out

  • No GitHub Releases published: versioning visible only on npm (@retaindb/sdk v5.1.1, @retaindb/local v0.2.1, unchanged as of 2026-07-23); 41 stars and no public third-party case studies as of 2026-07-23 — early-stage maturity and maintenance-risk should be weighed for production use
  • Server package is Business Source License 1.1 — internal self-hosting is allowed, but offering a competing hosted service is restricted; SDK is MIT, local and MCP packages are Apache-2.0
  • Self-reported benchmark claims (88% preference recall stated as SOTA on LongMemEval, 13ms retrieval latency, 0% hallucination on 16 doc questions) lack publicly reproduced verification; treat as vendor-stated until independently confirmed

Best for

  • Coding-agent workflows: single-machine persistent memory for Codex, Claude Code, or OpenCode via MCP or CLI, no cloud account required
  • Agents needing multi-modal recall — exact-string hits (codes, IDs, error strings) that semantic-only search misses, plus preference, correction, and decision memory across sessions
  • Product teams wanting a self-hosted or cloud memory API with per-user isolation, session ingestion, and knowledge-source connectors (GitHub, Notion, Slack, PDF, arXiv, npm/PyPI docs)

How it integrates

Benchmark results

BenchmarkValueBackboneTrustSource
longmemeval79 accuracygpt-5.4Self-reportedRetainDB

Sources

Last verified 2026-08-01 · updated by refresh-framework-cards