TencentDB Agent Memory
github.com/tencentcloud/tencentdb-agent-memory- Category
- AI Agents
- Rank
- No. 721Tools index
- Listed in
- #4 Give an agent memory
- Pricing
- Open Source
- Platform
- web · cli
- Type
- TOOL
- Builder
- tencentcloud
- GitHub
- 27.0k stars
- Latest release
- v2.0.2-beta.2
- Date
About
A self-hostable memory hub that turns AI agent conversations, documents, and codebases into four reusable, governed memory types—Chat Memory, Skills, Wiki, and CodeGraph—so teams of agents can share context and expertise instead of relearning it each session. It includes a control panel for managing ownership, versioning, and access (private/team/ACL) across multiple agent frameworks.
What it does
It records agent sessions, distills conversations through four layers from raw exchanges to personas, and recalls selected context into later prompts. Separate knowledge services map documents and source code, while a proxy accepts Anthropic and OpenAI compatible requests. Existing repositories, files, and conversation histories can seed the system.
Why it's ranked here
The design tackles memory as an operational system, not merely search. It combines extraction, retrieval, code relationships, document structure, assignment, and access policy. That breadth is useful, but the beta label, three-service deployment, and required LLM configuration make it better suited to committed teams than casual experiments.
What's good
The layered conversation model separates raw records, extracted facts, scenarios, and personas. Skills carry versions, resources, trigger boundaries, steps, and validation rules. CodeGraph tracks callers, callees, and impact paths instead of stopping at text matches. Private assets remain owner-only, while team and restricted modes support broader or precise sharing.
Tradeoffs
The default setup launches core, hub, and proxy services, and asks for two groups of LLM parameters. The proxy requires Node.js 22 and contains several optional integrations, storage modes, routing paths, and compatibility routes. That creates more operational surface than a small memory library. Team Memory is explicitly described as beta and evolving quickly.
How to use it well
Use it when several agents repeatedly work against the same conversations, documents, procedures, and codebase. Import existing material first, review extracted assets, then assign only relevant memory to each agent role. It fits teams that value governed reuse across sessions. It does not run the agent loop or replace an agent framework.
Technical notes+
MemoryCore/index.ts is an OpenClaw plugin shell around the host-neutral TdaiCore, with local and client adapter modes, CLI registration, hook handling, capture, recall, and layered profile processing. MemoryCore/tsdown.config.ts builds an ESM Node entry while keeping declared dependencies and OpenClaw external. MemoryKnowledge/tsdown.config.ts builds both HTTP and MCP server entries; MemoryKnowledge/src/server.ts exposes Hono routes under /v3, Swagger documentation, Wiki, CodeGraph, tool discovery, LLM bindings, and auto-sync controls. MemoryKnowledge/drizzle.config.ts configures SQLite through Drizzle. MemoryProxy/src/server.ts accepts Anthropic Messages and OpenAI-compatible chat requests, plus bridge, administration, session, health, and compatibility routes. MemoryCore/vitest.config.ts, MemoryPanel/vitest.config.ts, and MemoryProxy/vitest.config.ts define component-specific Vitest suites.
Observed
- License
- MIT
- Primary language
- TypeScript, compiled as ESM for Node.js
- Install surface
- Repository deployment starts memory-core, memory-hub, and proxy together; an npm package is also published
- Runtime requirement
- Node.js 22; the README states 22.16 or newer, while the proxy enforces the 22.x line
- Interfaces
- OpenClaw plugin, CLI commands, HTTP APIs, Anthropic and OpenAI-compatible proxy routes, and an MCP server entry
- Storage
- The knowledge service uses SQLite through Drizzle; proxy storage can request COS and report degraded fallback
- Testing structure
- Separate Vitest configurations exist for MemoryCore, MemoryPanel, and MemoryProxy
Read from README.md, MemoryCore/index.ts, MemoryCore/kubb.config.ts, MemoryCore/tsdown.config.ts, MemoryCore/vitest.config.ts, MemoryPanel/vitest.config.ts, MemoryProxy/vitest.config.ts, MemoryKnowledge/tsdown.config.ts, MemoryKnowledge/drizzle.config.ts, MemoryPanel/src/index.ts, MemoryProxy/src/index.ts, MemoryProxy/src/server.ts, MemoryCore/bin/seed-v2.mjs, MemoryKnowledge/src/server.ts, MemoryKnowledge/bin/server.mjs.
What it can do
Extract chat memory from AI agent conversations
Agent conversation logs → Chat Memory records
Extract reusable skills from documents and codebases
Documents or codebase → Skill records
Generate a wiki from documents and conversations
Documents/conversations → Wiki entries
Build a code graph from codebases
Codebase → CodeGraph
Manage access control for memory assets (private/team/ACL)
Memory asset and permission settings → Access-controlled memory asset
Version memory assets
Memory asset → Versioned memory asset
Tags
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.