AgentMemory
github.com/rohitg00/agentmemory- Category
- AI Agents
- Rank
- No. 118Tools index
- Listed in
- #2 Give an agent memory
- Pricing
- Open Source
- Type
- TOOL
- Use case
- Agent Building
- Interfaces
- CLI · SDK · API
- Builder
- rohitg00
- GitHub
- 28.8k stars
- Latest release
- v0.9.29
- Date
About
Persistent memory system for AI coding agents that remembers context across sessions. Eliminates the need to re-explain architectures, bugs, and preferences by automatically capturing and compressing agent interactions into searchable memory.
What it does
AgentMemory turns coding activity into structured sessions, observations, summaries, and typed memories. Agents reach the shared store through hooks, MCP, or HTTP. Retrieval can combine keyword and vector matching, while optional model providers handle compression, summaries, image descriptions, and knowledge-graph extraction.
Why it's ranked here
The integration breadth is convincing: one service supports many coding agents through standard interfaces, with explicit connection, health, diagnostics, import, and removal commands. The main reservation is operational coupling. It requires a compatible iii engine, currently pins one engine release, and offers a rougher native Windows path.
What's good
It remains useful without a model API key by falling back to keyword search and on-device embeddings. Memory records carry project, session, file, concept, confidence, relation, and lifecycle fields. The viewer defaults to loopback access, checks host headers, applies a restrictive content policy, and requires bearer authentication when exposed beyond loopback.
Tradeoffs
Installation requires Node 20 or newer and a separately managed iii engine with a tightly pinned protocol version. Native Windows setup is manual, and the connection command is unsupported there, making WSL2 the preferred route. Rich compression and summaries require a configured model provider, which introduces credentials, provider behavior, and potentially recurring usage costs.
How to use it well
Use it when several agent clients revisit the same long-running codebase and need a shared, queryable record of decisions, failures, files, and conventions. Start with the zero-model mode, verify recall using the demo and diagnostics, then add compression only if summaries justify the cost. It does not replace the coding agent or the model provider.
Technical notes+
package.json defines an ESM TypeScript package, a library export, an npm CLI, Node 20 minimum, Apache-2.0 licensing, and iii-sdk 0.11.2. src/index.ts registers the iii worker, state, search, memory functions, HTTP triggers, MCP endpoints, telemetry, and viewer. src/config.ts selects model and embedding providers, supports a no-model fallback, and derives service ports. src/mcp/server.ts maps authenticated MCP requests onto internal memory operations. src/auth.ts uses HMAC-backed constant-time comparisons and creates viewer nonces. src/viewer/server.ts defaults to loopback, validates Host headers, restricts non-loopback startup, and proxies authorized requests to the REST service. src/cli-data-dir.ts resolves platform-specific storage and isolated instance directories.
Observed
- License
- Apache-2.0
- Primary language
- TypeScript, compiled and published as ECMAScript modules
- Packaging
- npm package @agentmemory/agentmemory, installable globally or runnable through npx
- Interfaces
- Command-line application, ESM library export, MCP server, REST API, hooks, and local web viewer
- Runtime
- Node.js 20 or newer with iii-sdk and a pinned iii engine protocol version
- Platform support
- macOS and Linux are supported directly; WSL2 is preferred on Windows, where native connection setup is unsupported
Read from README.md, package.json, src/cli.ts, src/index.ts, src/auth.ts, src/types.ts, src/config.ts, src/logger.ts, src/version.ts, src/cli-data-dir.ts, src/mcp/server.ts, src/viewer/server.ts.
What it can do
Capture AI agent interactions automatically
AI coding agent conversations and activities → Stored interaction data
Compress agent context into searchable memory
Raw agent interaction data → Compressed, searchable memory entries
Retrieve relevant context from previous sessions
Current session context or search query → Relevant historical context and information
Remember software architectures across sessions
Architecture discussions and explanations → Persistent architecture knowledge
Store and recall bug information
Bug reports, fixes, and debugging sessions → Historical bug context and solutions
Maintain user coding preferences
User preference data from coding sessions → Persistent preference settings
Integrate with AI coding platforms via APIs
Platform-specific hooks and API calls → Cross-platform memory synchronization
Tags
Tech Stack
Comments (0)
No comments yet
Editorially curated, with community endorsements as a secondary signal. Corrections welcome.