Capture the knowledge inside the work.
Our background observer turns supported agent activity into observations and session summaries. Keep useful decisions and discoveries available while you focus on the task.
See automatic captureClaude-Mem vs Memori
A useful discovery should outlast the task that produced it. We capture decisions, discoveries and fixes from supported agent activity, then make that knowledge searchable in later sessions. Follow the timeline behind a result and give your next run the context you already earned.
Source review:

Production experience
Alessandro ran our open-source engine on three machines for 70 days. His customized deployment captured 22,718 observations, including 1,262 bug fixes and 757 decisions. Explore the fleet, the search index and the data he published.
Read the production case studyAI memory for agents
Our background observer turns supported agent activity into observations and session summaries. Keep useful decisions and discoveries available while you focus on the task.
See automatic captureWe connect compact search results to their timeline and full observations. Recover the context around a past decision and put that evidence in front of your agent.
Explore source-backed recallStart with our open-source engine in supported agents. Add CMEM Pro to sync eligible memory across devices and retrieve it from connected MCP clients.
Connect your agentsMemori turns agent execution into structured memory through its gateway and SDK workflows. We put a capture-and-recall loop into supported agents, with observations and a timeline that keep the evidence behind useful knowledge within reach.
Why choose Claude-Mem
Choose us to give your next task the benefit of what you already learned. Install supported agent connections, capture useful context in the background, then search the observations and inspect their timeline. Add CMEM Pro to carry eligible memory across connected tools and devices.
Our observer turns supported agent activity into searchable observations and session summaries. Build a memory of decisions and fixes that your next session can revisit.
Follow a compact search result into the surrounding timeline and full observation. Give your agent the evidence behind a decision, with optional sync to bring that context into connected tools.
Side by side
From what gets saved to what comes back in the next session. Each detail below links to its documentation.
| Workflow | Claude-Mem | Memori |
|---|---|---|
| Capturing agent work | Supported agent hooks feed a worker that turns activity into structured observations and session summaries. Claude-Mem architecture | A registered SDK client captures interactions automatically. Memori also documents execution traces including tool calls, decisions and outcomes, with structured primitives and rolling summaries. Memori TypeScript BYODB quickstart · Memori agent trace and execution |
| Inspecting remembered context | Session context plus a search → timeline → full-observation workflow, so the agent can inspect the notes behind an answer. Claude-Mem progressive disclosure | Automatic recall injects relevant memories before a wrapped LLM call. Explicit recall returns facts with identifiers and ranking information for custom prompts, interfaces or debugging. How Memori works |
| Storage choices | The open-source engine stores memory in local SQLite with full-text search. Optional Chroma adds semantic search; observation uses your configured model provider. Claude-Mem architecture · Claude-Mem installation | BYODB supports SQLite and other documented database engines; the database stores conversations and structured memory. Memori Cloud manages storage separately. Database ownership does not by itself establish where augmentation runs. Memori BYODB overview · Memori TypeScript BYODB quickstart |
| Coding-client connection | The installer configures supported agent integrations. Capture and recall depend on the selected integration; verify saved observations and a fresh-session query. Claude-Mem installation | The Cloud MCP guide connects Claude Code, Cursor, Codex and other clients with an API key and attribution headers. Tools include recall, summary, compaction and augmentation; connecting MCP alone is distinct from wrapping every application LLM call. Memori MCP client setup |
| Memory scope | Optional CMEM Pro sync moves eligible memory records between your devices and makes synced notes available through your private MCP connection. Capture still needs a supported integration. Claude-Mem cloud sync | Memories are attributed to an entity, process and session. Consistent identities matter when reusing context across applications; MCP clients can scope attribution per workspace. How Memori works · Memori MCP client setup |
| Application setup | Run the installer on your computer, select your agents and memory provider, then verify a captured observation and a recall query. Claude-Mem installation | The TypeScript BYODB quickstart registers an LLM client, supplies a database connection, builds storage tables and sets attribution. Short-lived scripts wait for asynchronous augmentation before testing recall. Memori TypeScript BYODB quickstart |
Try it on real work
Start with one real task. See what we capture, then put that memory to work in a fresh session.
Run our installer, choose your supported agent and memory provider, and finish setup. The installation guide walks you through the connection.
Work through a task with a decision worth remembering. Our observer turns supported agent activity into observations you can inspect.
Start a fresh session and ask about that decision. Search the memory, follow its timeline and open the supporting observation. Add CMEM Pro when you want synced memory in connected tools.
The installer guides your agent and provider choices. Then check the captured notes and ask a recall question.
npx claude-mem installQuestions
Yes. Its execution-memory documentation includes tool calls, decisions, workflow steps and outcomes, stored as structured memory primitives and rolling summaries. The comparison is about the workflow you want to operate, not exclusive ownership of execution memory.
Yes. Its BYODB documentation includes SQLite and other engines, and its TypeScript quickstart demonstrates a SQLite connection. Review the database adapter and augmentation configuration for your deployment.
Choose Claude-Mem when you want supported agent capture and a session-history workflow available through an installer. Search, inspect the surrounding timeline and open the full observation to understand what happened. That is a workflow fit, not proof of higher accuracy than Memori.
No. Compare the same tasks and models, verify saved context and fresh-session answers, and count capture, augmentation, retrieval, storage and model costs. Vendor benchmark results are not a matched result for your project.
Check the details
Written by Claude-Mem from the documentation linked below. Product documentation was reviewed on 18 November 2026; this is not a hands-on benchmark or a promise that every deployment has the same features.