Knowledge Capture & Curation Agent
Extracts candidate facts with source anchors from any run or document, resolves entities, dedupes, detects contradictions, and writes to org or account memory under the write policy. Never writes an unsourced fact.
How it works
Ask it like
Model & guardrails
- Runs on MYAM shared · open 7B (Qwen) by default — you can point it at your own model.
- Autonomy: Runs end-to-end; a human approves any outward action. Nothing leaves without a human’s approval.
- Data: Internal data. Provenance is attached to every output.
Outcome
Fills in as real usage proves it — from a Target we set, to Eval-measured (passes our test suite), to Platform-measured (across ≥20 accounts), to Verified (customer outcomes).
Also tracked: Contradiction-detection recall ≥ 0.8 · Unsourced writes = 0
Works well with
Executive Digest Agent
For teams & operators
A scheduled, length-capped digest of what changed, what needs a decision, and what is at risk across every fleet the org uses, with every line linked to its source artifact or record.
Moves: Decision latency on surfaced items
Meeting Notes & Productivity Agent
For teams & operators
Turns a transcript into structured notes, owner-assigned action items, a ready-to-send follow-up email, and a living thread document. Map-reduce over long transcripts with no silent truncation; due dates verbatim or null.
Moves: Action-item recall
Process Runbook / SOP Agent
For teams & operators
Converts tribal knowledge into runbooks with prerequisites, steps, roles, systems, validations, and exception paths; labels inferred steps; flags gaps for the process owner; schedules staleness review.
Moves: Runbook acceptance