A two-tier memory API server for AI agents built on Redis. Working memory is session-scoped and fast; long-term memory is persistent and searchable across sessions. Exposes both a REST API and a Model Context Protocol (MCP) server from the same backend, so any MCP-capable agent or HTTP client can connect without code changes. Memory extraction strategy (discrete facts, conversation summary, user preferences, or custom) is configurable per deployment.
- Storage
- Redis is the sole storage backend — both tiers write to Redis data structures. Working memory uses ephemeral session keys; long-term memory uses persistent records with vector embeddings for semantic search. A pluggable memory vector database factory system allows swapping the vector store without changing the API surface.
- Retrieval
- Three complementary retrieval modes: vector-based semantic similarity, full-text keyword search, and combined hybrid search with metadata filtering. Retrieved memories are injected into agent context via the MCP tools or REST endpoints. A separate async task worker handles background memory consolidation jobs to keep long-term memory clean.
- Self-host
- Self-host: moderate
- License
- Apache-2.0
- Pricing
- Open-source Apache-2.0, free to self-host. Requires a running Redis instance (Docker Compose setup provided). Published on Docker Hub (`redislabs/agent-memory-server`). No hosted cloud tier found. · Free / OSS
- GitHub stars
- 304
- Last release
- 2026-04-10
- Last commit
- 2026-07-09
- First catalogued
- 2026-06-28
Strengths
- Dual interface (REST + MCP) from a single backend — agents can use whichever protocol fits their stack without schema changes
- Two-tier design cleanly separates session-scoped working memory from persistent long-term memory
- Maintained by Redis, Inc. — strong operational backing; Docker-compose quickstart included
- Configurable extraction strategies (discrete, summary, preferences, custom) let operators tune memory behavior per use-case
Watch out
- Redis dependency: all memory lives in Redis; if Redis goes down, both tiers are unavailable — production deployments need Redis HA
- License is Apache-2.0 per LICENSE file content; GitHub API reports NOASSERTION for the SPDX field — verify before legal review
- 295 stars (as of 2026-07-23); no independent benchmark results; no public case studies yet
- `pip install agent-memory-client` is a thin HTTP client, not an in-process memory library: every add/search call is a network request to the agent-memory-server process, which itself requires a running Redis instance. Nothing runs inside your Python process — budget for two additional services (API server + Redis), not zero.
- Background task worker must run alongside the API server for consolidation jobs — two-process deployment adds operational overhead
Best for
- Agents already running in Redis-backed infrastructure that want persistent memory without adding a new database
- Teams wanting a single memory server accessible from both HTTP clients and MCP-native agents
How it integrates
Benchmark results
No sourced results yet.
Sources
- Monthly card re-verification — GitHub API repo metadata: 299 stars, GitHub reports no auto-detectable SPDX license (card license recorded from the repo's own LICENSE file), archived=false, last commit on the default branch 2026-07-09 (third-party)
- Redis Agent Memory Server README (vendor)
- GitHub API repo metadata (stars, Apache-2.0 confirmed via LICENSE file, server/v0.15.2 release) (third-party)
- Redis Agent Memory Server MCP docs — stdio mode (`uvx --from agent-memory-server agent-memory mcp`) and SSE/remote-http mode (`agent-memory mcp --mode sse --port 9000`); official Redis project confirmed (vendor)
- Redis Agent Memory Server REST API reference — `/v1/` endpoint catalog; working memory, long-term memory, memory-prompt, summary views, tasks; Bearer token auth; API version 0.14.0 (vendor)
- agent-memory-client PyPI page — Python client SDK (`pip install agent-memory-client`); async-first; LangChain integration; maintained by Agent Memory Server Team (redis org); v0.14.0 released 2026-02-06 (vendor)
- GitHub API re-check — 286 stars, pushed 2026-06-30; latest tag still server/v0.15.2 (2026-04-10), no new release (third-party)
- PyPI JSON API re-check — still v0.14.0 (2026-02-06); dependencies are httpx/pydantic/python-ulid only (no Redis driver), confirming the client is a pure HTTP wrapper with no in-process storage (third-party)
- GitHub API re-check — 295 stars, still license NOASSERTION per API (LICENSE file confirmed verbatim Apache-2.0 text, Copyright 2025 Redis, Inc.), latest tag still server/v0.15.2 (2026-04-10) (third-party)
- MCP docs re-check — remote mode is still documented only via SSE (`http://localhost:9050/sse`), no Streamable HTTP transport; card correctly omits remote-http (vendor)
Last verified 2026-08-01 · updated by refresh-framework-cards