Own the agent loop.
Record executions, recall applicable memory, request bounded influence, and submit outcomes with typed contracts.
→Install the SDK, connect MCP, and let a fresh session reuse verified experience without replaying raw history or expanding its power.
$ npm install engram-agent-surface // MCP client config { "command": "npx", "args": ["--yes", "engram-agent-surface@1.0.2"] } // one canonical lifecycle record → recall → influence → outcome
Record executions, recall applicable memory, request bounded influence, and submit outcomes with typed contracts.
→Connect a tool-using agent to the same lifecycle through a local installable MCP command.
→Use the OpenAPI boundary from a worker, service, or language-independent application.
→Capture what happened, the outcome, interpretation, and evidence.
The agent actively requests a bounded MemorySlice.
Engram names exactly what the memory may influence.
The result creates the next immutable memory version.
Copy the usage guide for an agent: when to record, when to recall, how to cite a slice, and when to stop on refusal.
Local MCP uses local Sibyl persistence. Hosted REST uses the separately deployed Wrangler Worker and D1. Choose the mode intentionally.
Clean commits, complete setup instructions, runnable proofs, SDK, REST, MCP, and canonical evidence are available in the public repository.
OPEN REPOSITORY ↗Start with the problem, follow the Engram lifecycle, and watch a fresh Session B request memory and change its provider decision.
WATCH REPLAY ↗Builder: Jason / Jaydearcadian.
Partner stacks: no Base or Virtuals usage is claimed or independently verified in this submission.
Engram persists completed executions through Sibyl, recalls an eligible MemorySlice, applies a bounded InfluenceGrant, and evaluates the outcome into an immutable update.
READ THE USAGE NOTE ↗