Skip to content

AI MEMORY FOR AGENTS · ELIZAOS

Give elizaOS knowledge it can find again.

We keep useful project knowledge available beyond the moment it is created. Save a selected note with our local worker, then use elizaOS’s existing MCP service to search its context and retrieve the original record.

Start with our explicit save-and-recall example. No model key or CMEM Pro account is needed for this example.

KNOWLEDGE THAT LASTS

Keep useful knowledge within reach.

Keep the decision, not another retelling.

We save the note your application chooses and return its record ID, so later recall can return to the original.

Find context through the service you already use.

Our example uses elizaOS’s existing MCP service to search, inspect nearby context and retrieve the saved note.

Keep project knowledge together.

We pass your selected project through each recall stage, including when the MCP child starts inside a worktree.

ILLUSTRATIVE NOTE · ACTUAL API AND MCP SEQUENCE

One decision. A record you can return to.

“A restore must reject snapshots missing required worker tables.”

The example saves this note through the worker API, then uses elizaOS’s MCP service to find context and retrieve the original. Your application chooses what to save.

  1. 01

    Save the chosen note

    Your application sends the selected note to the local worker’s JSON save API.

  2. 02

    Keep the receipt

    The worker confirms the observation ID before recall begins.

  3. 03

    Find project context

    elizaOS’s MCP service searches the selected project and reads the saved record’s timeline.

  4. 04

    Read the original

    Retrieve the confirmed observation by ID and check that it contains the expected note.

AI memory in production

70 days of work. One searchable memory.

Alessandro used our open-source engine across three machines for 70 days. His customized deployment captured the decisions, fixes and discoveries behind the work—and published the data.

Observations captured
22,718
Bug fixes recorded
1,262
Decisions recorded
757

See how he made that knowledge searchable. Then give your own agents a memory to build on.

Claude-Mem’s mascot files a note in a shared card catalogue connected to three computers.
Alessandro’s published field study · March–May 2026

START WITH ONE PROJECT

Save one note. Read the original back.

  1. Build Claude-Mem and start your local worker. Follow the installation guide. From a source checkout, inspect npm run worker:status and use npm run worker:start if needed. Note your worker’s actual port.
  2. Install the example’s pinned dependencies. Use Node.js and the existing Claude-Mem build prerequisites. The example keeps elizaOS core 1.7.2 with MCP plugin 1.7.0.
  3. Choose the project, bundle and worker. Run below from the source checkout. Replace YOUR_WORKER_PORT with the actual local port; use an absolute MCP bundle path when running elsewhere.
npm run build
npm install --prefix examples/elizaos
node examples/elizaos/memory.mjs \
  --project eliza-practice \
  --server "$PWD/plugin/scripts/mcp-server.cjs" \
  --worker http://127.0.0.1:YOUR_WORKER_PORT

Check the result. The program prints the confirmed save ID before search, timeline and get_observations output. The retrieved record must contain the example note.

The example’s note is defined in its text constant; inspect and adapt that value for your application. It has no --note option. Each invocation requests a save. To read existing knowledge without saving again, retain and reuse the observation ID in your application.

Keep the same loopback worker origin for save and recall, and preserve the explicit project scope and error checks. The script sends no note to a model provider; your worker’s existing privacy and sync settings still apply.

Before you connect

Does it record every conversation?

No. The example saves its selected synthetic note. It adds no transcript hooks, IDE configuration, startup injection or autonomous conversation.

Do I need a model key or Pro?

The deterministic local example needs neither. It uses the actual elizaOS runtime and MCP service without a model call. Model-driven applications have their own provider requirements.

Which versions should I use?

The example pins @elizaos/core 1.7.2 and @elizaos/plugin-mcp 1.7.0, reported verified on macOS. Keep that pair for this setup; newer versions, Windows and Linux require separate validation.

Can I connect a remote worker?

This example accepts only a credential-free HTTP loopback origin: localhost, 127.0.0.1 or [::1], with no path, query or fragment. It passes that same origin to the MCP child so save and recall reach the same worker. Remote and authenticated use need a separately verified setup.

What happens if a save or recall fails?

The run fails on a save error, MCP error or missing expected note. A confirmed save ID is printed before recall, so a later failure does not mean the save was undone. Check existing memory before repeating an uncertain write. Each invocation requests a save; retain an existing ID for recall-only application paths.

How are projects kept separate?

Each read carries the selected project and an explicit one-item projects list. Preserve both fields when adapting the example. These filters keep worktree reads scoped; they are not multi-user authorization.

What happens when I disconnect?

The started service closes its MCP connection in finally. Stop using the example or remove only its own application configuration. Disconnecting preserves saved memories and other agents. Existing save privacy rules and worker synchronization settings still apply.

Build on a working example.

Source reviewed November 20, 2026. The published fixture starts an isolated worker, runs the actual elizaOS runtime and MCP service, checks the original text and project boundaries, and verifies that a stopped worker fails the save. The guide reports macOS verification. This is not a model-quality benchmark or an autonomous conversation test.

Keep the knowledge. Give the next step a stronger start.

Add an explicit memory path to your elizaOS application, with project context and a retrievable source behind each saved note.