Self-serve product
The Agent Memory Audit
We review your agent-knowledge estate against a playbook proven on a live system — and hand you the findings, the failure classes, and the restructure plan. Fully async.
Audit
$349 AUD
- One agent surface (~200 knowledge files)
- Findings + failure classes
- Quick wins
- Restructure plan (summary)
Audit+
$749 AUD
- Full estate, multi-agent
- Findings + failure classes
- Quick wins
- Full per-layer restructure plan
- Two written async Q&A rounds
How it works
- Submit — read-only repo grant or a bundle (CLAUDE.md / prompts, wiki export, RAG config, agent memory stores) plus a short questionnaire.
- We audit — the loop runs your estate against the playbook; a review gate sits before anything leaves.
- Report in 48 hours — written, specific, yours.
Failure classes we score
- Facts with no freshness owner
- Absence read as fact
- Copies drifting from their source
- Presence read as relevance (other people's world in your model)
- Silent-clobber surfaces
- Convention drift across agents
- Trust pricing missing from the index
Boundaries
Read-only access, always. We'll sign your NDA. Your submission is deleted after delivery plus 14 days. Nothing you send us trains anything. If your bundle contains live secrets, we stop and tell you how to rotate them — that one's free.
Read the free build story first if you want the scars behind the method: Two brains, one truth.