Memorylayer

Hosted memory runtime for agents

Connection kit

Give an agent a workspace in minutes.

Create a workspace key, fetch the connection kit, load recent memory, then save a checkpoint when the session ends.

4

startup calls: bootstrap, recent, tools, context

60

hosted MCP tools available through one endpoint

2

starter skills downloadable as JSON or Markdown

key config recall handoff
01 / Key
Create one scoped key

Use a separate workspace API key for each agent, worker, or integration so usage and revocation stay clean.

02 / Config
Fetch the connection kit

`/connect` returns endpoint URLs, MCP settings, starter skills, and the startup calls a client should run.

03 / Recall
Load working context

Start with recent memory, then ask `recall_context` for compact task-specific state before making changes.

04 / Handoff
Save the stop point

End with `session_checkpoint` or `session_handoff` so the next agent session starts with the right state.

Config JSON`/api/workspaces/{slug}/connect` returns the full client profile, including startup calls.
Env output`/api/workspaces/{slug}/env` returns copyable variables for local launchers.
Runtime bridge`/mcp` stays the single bridge for higher-level memory tools.

Boot sequence

This is the recommended order for a new agent session.

1. Bootstrap

Load endpoint URLs, tool discovery, starter skills, and auth shape.

2. Recall

Fetch recent memory and compact task context before changing files.

3. Work

Use direct routes for search/write and the MCP bridge for higher-level tools.

4. Handoff

Save `session_checkpoint` or `session_handoff` so the next session resumes cleanly.

Client config

Replace `{slug}` with the workspace slug and authenticate with a workspace API key.

JSON profile

curl -H "Authorization: Bearer $MEMORYLAYER_KEY" \ "$MEMORYLAYER_URL/api/workspaces/$SLUG/connect"

Env block

curl -H "Authorization: Bearer $MEMORYLAYER_KEY" \ "$MEMORYLAYER_URL/api/workspaces/$SLUG/env"

Recommended calls

Keep this small. The goal is enough state to start, not a giant prompt dump.

Load compact context

Ask the hosted bridge for task-specific context after bootstrapping.

curl -X POST -H "Authorization: Bearer $MEMORYLAYER_KEY" -H "Content-Type: application/json" \ -d '{"tool":"recall_context","args":{"query":"current task","max_tokens":1200}}' \ "$MEMORYLAYER_URL/api/workspaces/$SLUG/mcp"

Save the stop point

A checkpoint is cheap and makes the next session faster.

curl -X POST -H "Authorization: Bearer $MEMORYLAYER_KEY" -H "Content-Type: application/json" \ -d '{"tool":"session_checkpoint","args":{"note":"Finished current task; next step is documented","limit":8}}' \ "$MEMORYLAYER_URL/api/workspaces/$SLUG/mcp"