# TencentDB Agent Memory (slim)

> TencentDB Agent Memory is Tencent Cloud's MIT-licensed, self-hosted memory stack for coding agents. A proxy injects layered conversation memory, skills, wiki pages and code graphs into each request, with a team panel and HTTP API over SQLite.

- Full: https://www.anchorterminal.com/tools/tencentdb-agent-memory.md (~10,700 tokens) · this version ~2,730 tokens · JSON https://www.anchorterminal.com/tools/tencentdb-agent-memory.json · canonical https://www.anchorterminal.com/tools/tencentdb-agent-memory
- Index: https://www.anchorterminal.com/llms.txt · API: https://www.anchorterminal.com/api/v1/index.json · MCP: https://www.anchorterminal.com/mcp · Updated: 2026-10-30 · pricing https://www.anchorterminal.com/tools/tencentdb-agent-memory#pricing · reviews https://www.anchorterminal.com/tools/tencentdb-agent-memory#reviews
- Licence: CC BY 4.0, credit Anchor Terminal and link this page (https://www.anchorterminal.com/terms/#data)

**C · 56/100 · rank #767 of 1239 · #14 in Agent memory · not agent-ready · confidence medium**


Assessment: A self-hosted team memory stack with layered recall, per-asset visibility and ACLs, documented HTTP APIs and frequent releases, 2.0.0 to 2.0.3 since August. The default deployment publishes its ports on all host interfaces with the core's bearer gate off, data-plane deletes trust a service-ID header, and the proxy's rate-limit and session admin routes have no authentication.

## Facts

- Kind: HTTP API · vendor: Tencent Cloud · category: Agent memory · legal entity: Tencent · provenance 75/100
- Local only (HTTP): oci `agentmemory/memory-core`, oci `agentmemory/memory-hub`, oci `agentmemory/memory-proxy`, npm `@tencentdb-agent-memory/memory-tencentdb`, npm `@tencentdb-agent-memory/memory-sdk-ts-v2`
- Auth: API key · pricing: Free · x402: no · licence: MIT
- Probe metrics: not measured yet (probes haven't run)
- Surface graded: The default branch (master) at 2.0.3, read from raw files on 30 October 2026. Memory Core (port 8420, 109 `/v3` endpoints), the Memory Hub panel and Knowledge service (8125 and 8424) and the Memory Proxy (8096), run from Docker Hub images by the scripts in deploy/global-images. Nothing was installed or run
- Age: The OpenClaw plugin reached npm as 0.1.0 on 31 March 2026 (its changelog dates 0.1.0 to 25 March). The 2.x Hub line's first public release, 2.0.0-beta.1, is dated 21 July 2026 and 2.0.0 3 August 2026. The repository's creation date wasn't read
- How agents connect: Through the proxy, by pointing an agent's Anthropic or OpenAI base URL at it. Each turn it injects the bound agent's L2 and L3 memory, matched skills and wiki or code-graph context into the system prompt and forwards to the configured upstream model. Guides cover Claude Code, CodeBuddy, WorkBuddy, Codex, DeepSeek Harness, OpenCode, Hermes, OpenClaw and Pi
- Session binding: On a new session the proxy asks the user to pick a team, agent and task through the client's AskUserQuestion tool. Hermes and OpenClaw can't answer the form, so they need `x-team-id`, `x-agent-id`, `x-task-id` and a static `x-conversation-id` in their config, or the proxy passes the session through without memory
- Memory layers: L0 raw conversation, L1 atoms (episodic, persona and instruction), L2 scenario blocks and L3 core persona. Background workers extract each layer from the one below as thresholds are hit. Recall uses BM25, vectors and reciprocal rank fusion, capped by item count, character budget and timeout
- Other assets: Skills extracted from conversations and tool calls, with versions and resource files. Wiki pages built from imported documents, and code graphs of symbols, callers and impact paths built from repositories. Knowledge exposes 7 read-only wiki tools and 9 code-graph tools through `/v3/tools/list` and `/v3/tools/call`
- Storage: SQLite with sqlite-vec by default, on the Memory Core container volume. MongoDB is an experimental backend, off by default, and switching backends doesn't migrate data. The plugin line also supports Tencent Cloud Vector Database (TCVDB)
- Access control: Assets are `private` by default, or `team`, `restricted` (user, role and agent ACLs) or `agent`. A global System Admin and team Admin and Member roles. Per-user `sk-mem-` keys, several per user, revocable except the last. Data-plane routes take team, agent and user IDs from the body or headers and fall back to a `default` bucket
- Unauthenticated routes: Per the API references, Memory Core's bearer gate is off when its key is empty (the default), data-plane deletes and `/v3/chat-memory/clear` make no user-level check and so rest on `x-tdai-service-id` alone, and the proxy's `/v3/admin/rate-limits` and `/v3/session/*` routes have no authentication. Knowledge leaves `tools/*` and read endpoints open even with its service key set
- Deploy defaults: The start scripts publish ports 8420, 8096, 8125 and 8424 with Docker's `-p` and no host address, which binds all interfaces, and set the gateway host to 0.0.0.0. The deploy README says its default admin key and gateway key are only for local trials and must be replaced before exposure
- OpenClaw plugin: `@tencentdb-agent-memory/memory-tencentdb` (npm 1.0.3, 22 September 2026), a four-layer local memory plugin with `tdai_memory_search` and `tdai_conversation_search` tools, short-term context offload to Mermaid canvases, and a Hermes gateway on port 8420. Its npm postinstall runs a bash script that locates OpenClaw and patches its dist files with sed, keeping .bak copies
- SDKs: TypeScript `@tencentdb-agent-memory/memory-sdk-ts-v2` (npm latest 1.0.0-beta.1, 1.0.1-beta.1 on the beta tag, 14 August 2026) and a Python SDK whose pyproject names `tencentdb-agent-memory-sdk-python-v2` 1.0.1-beta.1, while the changelog says `pip install tencentdb-agent-memory-sdk-python`. PyPI wasn't readable
- What leaves the machine: Conversation text goes to the configured memory LLM for L1 to L3 extraction, skill and wiki building, and every proxied request goes to the configured upstream model. The Hermes Docker image on the main branch defaults to Tencent Cloud's LKE endpoint with DeepSeek-V3.2. Analytics export to ClickHouse and the plugin's metrics reporting are off by default
- Documentation language: README, INSTALL and ROADMAP in English and Chinese. CHANGELOG.md and the Memory Core, Knowledge, Proxy and Panel API references in Chinese. The Knowledge service also has an OpenAPI 3.0.3 file of 28 paths
- Licence: MIT, 'Copyright (C) 2026 Tencent', on both branches and in the npm metadata. The README credits code taken from colbymchenry/codegraph and from Nous Research's Hermes Agent
- Cost: Free software. LLM calls for extraction, wiki and code-graph building and the proxied model, and any MongoDB or ClickHouse server, are the owner's

- Scores: Reliability 51, Performance pending, Schema & documentation 61, Agent ergonomics 68, Security & auth 30, Payments & pricing 60, Task success pending, Maintenance & community 70, Transparency & trust 68 · total over the 7 assessed categories
- Why: Reliability, Local-software reading, as software the owner runs. · Schema & documentation, API reading for a self-hosted HTTP service. · Agent ergonomics, Read with the API lines. · Security & auth, Read as self-hosted software with a network API. · Payments & pricing, Read with the self-hosted rule. · Maintenance & community, Release 2.0.3 on 26 October 2026 by the changelog, with images tagged 1.0.3 on Docker Hub the same day, 4 days before this check (30). · Transparency & trust, MIT on both branches and in npm metadata, with the codegraph and Hermes Agent code it reuses credited in the README (30 of 30).
- Sources: 32, open questions: 10, both in the full twin
- Capabilities: memory.store, memory.search, memory.user, memory.delete, knowledge.search
- JSON: https://www.anchorterminal.com/api/v1/tools/tencentdb-agent-memory.json
- Verify (for the vendor): the badge `https://www.anchorterminal.com/badges/tencentdb-agent-memory.svg` or a link to https://www.anchorterminal.com/tools/tencentdb-agent-memory from a page on tencent.com or one of its subdomains, a page under github.com/tencentcloud, or the README of github.com/TencentCloud/TencentDB-Agent-Memory, then `POST https://www.anchorterminal.com/api/v1/verify` `{"slug", "url"}` or `verify_listing` at /mcp; re-checked weekly, no effect on the grade. Snippets in the full twin.
- Setup: needs a person once, about 30 minutes; no card; free: Free and open source under MIT. The memory LLM and the proxied upstream model are paid to whichever provider the person configures, or run on a local OpenAI-compatible server.. Ask your human: "Please run TencentDB Agent Memory (Tencent Cloud's free MIT memory stack) with Docker, set its gateway and knowledge keys, create a team, agent and task for me in the panel, and give me the proxy URL and a non-admin sk-mem- user key. Note that every proxied turn is stored and sent to the LLMs you configure."
- Support: Discord https://discord.gg/dJQM6mKMF; https://github.com/TencentCloud/TencentDB-Agent-Memory/issues; https://github.com/TencentCloud/TencentDB-Agent-Memory/discussions; more in the full twin
- For companies: data in self-hosted; EU residency not stated; trains on your data no; zero retention not stated; no certifications stated

## Before you call it

1. Set `MEMORY_CORE_GATEWAY_API_KEY`, `KNOWLEDGE_SERVICE_KEY` and a non-default admin key before exposing ports 8420, 8424, 8096 or 8125 beyond the host
2. In headless runs send `x-team-id`, `x-agent-id` and `x-task-id` headers to the proxy. Without all three it falls back to an interactive picker and skips memory
3. Expect every proxied turn to be stored. The proxy records upstream messages and model replies to Memory Core, and injected memory lands in your system prompt
4. Check `data.valid` on `/v3/meta/auth/verify` and `data.isError` on `/v3/tools/call`. Both can fail inside a success envelope
5. Retry `/v3/session/add` only on 500, 503 or a network error. The API reference says it isn't idempotent and writes a new object each call

## Connect

```bash
git clone https://github.com/TencentCloud/TencentDB-Agent-Memory.git
cd TencentDB-Agent-Memory/deploy/global-images
cp .env.example .env
./start-all.sh
```

```bash
export ANTHROPIC_BASE_URL=http://127.0.0.1:8096/claude-code/default
export ANTHROPIC_AUTH_TOKEN='<sk-mem- user key>'
claude --model <PROXY_UPSTREAM_MODEL>
```
Full config and headless snippets are in the full page. Through letme (picks today, calling later): https://letme.dev/tencentdb-agent-memory

## Similar tools

| Tool | Grade | Score | Shared capabilities | Slim |
| --- | --- | --- | --- | --- |
| Amazon Bedrock AgentCore Memory | A | 79.7 | memory.store, memory.search, memory.user, memory.delete | https://www.anchorterminal.com/tools/agentcore-memory.min.md |
| Zep | BB | 74 | memory.store, memory.search, memory.user, memory.delete | https://www.anchorterminal.com/tools/zep.min.md |
| Honcho | B | 64.2 | memory.store, memory.search, memory.user, memory.delete | https://www.anchorterminal.com/tools/honcho.min.md |
| Supermemory API + MCP | B | 63.8 | memory.store, memory.search, memory.user, memory.delete | https://www.anchorterminal.com/tools/supermemory.min.md |
| past.dev | B | 63.7 | memory.store, memory.search, memory.user, memory.delete | https://www.anchorterminal.com/tools/past-dev.min.md |


## Panel reviews (0, desk reviews from public material, no calls made)