ClariLayer

Every saved entry keeps its source and applicability visible. You see which conversation a decision came from, which project it applies to, and which version it references. When a plan changes, you correct the decision or forget it from the scope where it no longer belongs. That audit trail matters. An AI memory you can't inspect is hard to trust. Source: clarilayer.com.

ClariLayer documents four verbs, and two of them carry most of the daily work. Recall pulls relevant saved decisions, definitions, notes, and caveats into the current task. Remember saves a confirmed decision, constraint, definition, or correction for later sessions. Remember is an explicit write the agent triggers. It doesn't happen behind your back.

You can seed context from structured sources: SQL, dbt models, CLAUDE.md notes, a data dictionary or codebook, and semantic-layer models. That gives an agent the definitional grounding it needs before it touches an analysis. Setup is deliberate. You pick a source, a space, and an extraction provider, then review the exact proposal before accepting it.

Reconcile checks a compatible analytics SQL definition against warehouse actual_sample evidence, or a HubSpot CRM contract against row-free crm_evidence. When they don't match, the mismatch shows up as a caveat. The site calls this a check no notes file can make, and that's a fair claim. A caveat tied to live source evidence beats a stale comment every time.

Connection alone doesn't read your history or start capturing. You select a supported local history source, a space, and an extraction provider, then preview the batch before accepting it. Ongoing capture is enabled separately, with its own authorization and exclusions, so nothing starts reading your history until you explicitly say it can and review exactly what it would pull in. Here's the catch. Current qualified extraction uses a version-specific Claude CLI profile. OpenAI and Cursor extraction providers aren't available, and native Cursor history qualification remains limited.

ClariLayer connects as a remote MCP server over Streamable HTTP, so it slots into clients that support the protocol. The site lists Claude Code, Cursor, Codex, and claude.ai. Setup runs through each client's own configuration, and OAuth is offered where compatible, which means you keep your existing workflow instead of moving everything into a new app just to get persistent context.

The Governed Context Edge is a shared layer for team definitions. You propose a definition up to a shared canon, adopt the canon down with your own agency, and keep disagreement visible. It's in private pilot with no public price. Worth knowing: project context is proposed deliberately, and personal memory is never silently promoted into the team layer.