AI review and approved context

Use AI review when saved notes overlap or disagree. Memford prepares a source-based draft, keeps potential conflicts visible and leaves the decision with you. Original notes remain available.

Generate a draft

1. Open the project and choose AI review.
2. Read the processing information and allow AI processing of this project’s notes.
3. Choose Generate draft. Wait for the actual result before relying on a summary.
4. Inspect the proposed facts, hypotheses, decisions and conflicts alongside their source references.

Generation uses the active note snapshot. Source references help you check a claim; they are not proof that the model interpreted it correctly. The current input limit is 100 complete notes and 48 KiB per run. If the snapshot exceeds it, Memford rejects the run before sending the notes. It does not silently truncate them or automatically reduce them in chunks. This is a Memford input guard, not a Cloudflare model context-window specification.

Resolve disagreement at its source

If one agent records one decision and another records an incompatible decision, inspect both sources and make the intended decision explicit. Update the source notes or append the agreed outcome, then generate and inspect a fresh draft. The draft view supports approval or rejection, not editing the conflict directly into a new fact.

AI may overlook a contradiction or identify a false one. Do not treat an AI assessment as an automatic decision about the project.

Approve the working version

Approve a draft only when it represents the intended project context. Reject it otherwise. Memford checks whether covered source notes still match before accepting approval. Editing, deactivating or deleting a covered source invalidates its approved context; new notes remain separate until included in a later approved draft.

Authorized agents can request compact reads. These return valid approved context plus notes not covered by it, while preserving assigned standards in full and skill summaries. Ordinary reads continue to include the active original-note view. Compact reads can reduce repeated context, but no fixed token or cost saving is guaranteed.

Approval advances the project change revision. A supporting subscribed client receives a change hint, requests the current context and decides how to use it in its own session. See MCP resources.

Optional automatic drafts

Enable automatic drafts allows new notes to trigger preparation of a draft, at most once per hour per project and within shared service limits. This setting never approves or publishes a draft automatically. Disabling it stops future automatic processing; it does not erase existing drafts or undo a provider request already sent.

Processing and limits

The workspace identifies the processing provider and model before consent. Connected providers may process data outside the EU even though the project database is located in the EU. Read the Privacy Policy before enabling processing of sensitive project material. Source, schema and output checks can reject an unusable result; budget or capacity limits may also prevent generation.

Notes and history explains source management. Skill library covers reusable procedures, which remain separate from AI summaries.