What is new in TencentDB Agent Memory v2.0?

TencentDB Agent Memory v2.0 adds four production-ready features on top of the beta: Skill forced archiving so critical procedures cannot silently drift out, CodeGraph scheduled sync that keeps the code index current, bilingual switching in the admin panel, and expanded admin asset management for the four memory types.

Can Claude Code and Codex both use TencentDB Agent Memory?

Yes. The Memory Proxy in front of the hub exposes both Anthropic and OpenAI-compatible endpoints, so Claude Code, OpenAI Codex, and any agent that speaks either protocol can read from and write to the same shared memory pool without a bespoke SDK.

Is TencentDB Agent Memory open source?

Yes. TencentDB Agent Memory is released under the MIT license. Docker images ship for multiple architectures, and the ./start-all.sh script boots the full stack (memory server, proxy, admin panel, initial admin credentials) so a team can self-host without depending on any Tencent Cloud service.

What are the four memory asset types?

Chat Memory keeps user preferences and interaction history; Skill captures reusable workflows extracted from finished tasks; Wiki structures documentation as interlinked pages; and CodeGraph indexes code symbols, call relationships, and impact analysis so an agent can reason about a repository the same way a senior engineer would.

Tencent Cloud · 2026-08-03 · major

TencentDB Agent Memory v2.0 — team-level memory hub for AI coding agents

Tencent Cloud open-sources v2.0 of TencentDB Agent Memory, an MIT-licensed hub that turns team conversations, docs, and code into four reusable assets (Chat, Skill, Wiki, CodeGraph) shared across agents via Anthropic and OpenAI endpoints.

TencentDB Agent Memory GitHub repository social card

A shared memory layer that lets a team's coding agents keep learning from each other's work.

Key specs

License	MIT
GitHub stars	14.6K

Quick facts

Maker	Tencent Cloud
Version	v2.0.0
Released	2026-08-03
License	MIT
Memory types	Chat Memory, Skill, Wiki, CodeGraph
Endpoints	Anthropic and OpenAI-compatible
SDKs	TypeScript, Python

What is it?

TencentDB Agent Memory is a self-hostable server that stores four kinds of reusable memory for AI agents: Chat Memory (preferences and history), Skill (workflows harvested from past tasks), Wiki (structured docs with relationships), and CodeGraph (an indexed view of a code repo). Version 2.0.0, released 2026-08-03, is MIT-licensed on GitHub with 14.6K stars.

How does it work?

Raw agent conversations flow through layered extraction (L0–L3) that promotes them into the four asset types. A Memory Proxy in front of the hub speaks both Anthropic and OpenAI protocols, so Claude Code, Codex, and any OpenAI-compatible agent can pull from and write to the same team knowledge base without special SDKs.

Why does it matter?

Agent memory has been per-user and per-session; TencentDB Agent Memory pushes it to the team level so a workflow one developer's agent works out becomes reusable by the whole team's agents. v2.0 adds scheduled CodeGraph sync, forced skill archiving, and an admin dashboard, turning the tool from a demo into something a company can actually deploy.

Who is it for?

engineering teams running fleets of coding agents, self-hosted AI platform ops

Frequently asked questions

What is new in TencentDB Agent Memory v2.0?: TencentDB Agent Memory v2.0 adds four production-ready features on top of the beta: Skill forced archiving so critical procedures cannot silently drift out, CodeGraph scheduled sync that keeps the code index current, bilingual switching in the admin panel, and expanded admin asset management for the four memory types.
Can Claude Code and Codex both use TencentDB Agent Memory?: Yes. The Memory Proxy in front of the hub exposes both Anthropic and OpenAI-compatible endpoints, so Claude Code, OpenAI Codex, and any agent that speaks either protocol can read from and write to the same shared memory pool without a bespoke SDK.
Is TencentDB Agent Memory open source?: Yes. TencentDB Agent Memory is released under the MIT license. Docker images ship for multiple architectures, and the ./start-all.sh script boots the full stack (memory server, proxy, admin panel, initial admin credentials) so a team can self-host without depending on any Tencent Cloud service.
What are the four memory asset types?: Chat Memory keeps user preferences and interaction history; Skill captures reusable workflows extracted from finished tasks; Wiki structures documentation as interlinked pages; and CodeGraph indexes code symbols, call relationships, and impact analysis so an agent can reason about a repository the same way a senior engineer would.

Try it

git clone https://github.com/TencentCloud/TencentDB-Agent-Memory && ./start-all.sh