The job
Use cases
Same work. Whichever agent. One truth.
Each use case is one specific job, the friction you have today, the workflow with Unforgetter, the scopes it uses and where the product honestly stands. No composite customer stories, no invented numbers.
Pick the job
Five jobs that outlive a single session.
Founders · switching agents
Switch agents without re-explaining the company.
Strategy in one tool, code in another, outreach in a third. Each one is good; none of them knows what the others were told.
Today
Where it breaks
You paste last month's update, the tone notes and the list of things that must not be promised. The new agent still asks what the tone should be. Corrections you made in one tool never reach the other.
With Unforgetter
What changes
Your writing rules, the customer-copy policy and the last four updates are compiled together. The agent opens by drafting in your voice and states what is unfinished, because that is your rule.
Decision
Product updates always state what is unfinished
14 Aug · your rule after an update overpromised
Method
Plain sentences, no superlatives, no invented numbers
An instruction-only skill, versioned and portable
Open thread
Pricing is not public
Customer copy must not imply a price
Scope
Customer-facing product material only
Nothing internal or financial enters customer-facing copy
Provenance
5 sources, freshest 21 Aug
Including the four updates the tone is drawn from
The second agent opens with: "Drafting the product update. Plain sentences, no superlatives, and it says what is still unfinished, the way your last four updates read."
Scopes and data used
Customer-facing product material
Writing method and customer-copy policy. Internal finance excluded by default.
Expected effect
What you should notice
Fewer tone rewrites and fewer repeated instructions. Draft corrections can be compared across agents without making voice parity a claim.
StatusWorking now
Compatible across your agents
The same scoped memory can serve Claude, ChatGPT / Codex, Cursor and Grok Bot through MCP. Voice parity between agents is being measured, not yet claimed.
Builders · resuming long work
Resume long-running work where it actually stopped.
A feature spans weeks and several sessions. What was decided in week one is buried under what was discussed in week three.
The job
Extend the booking flow so a late cancellation is charged against the right date.
Today
Where it breaks
The agent asks which date to measure against. That was decided three weeks ago, but the decision is in a chat you closed. It guesses, ships the wrong rule and review catches it late.
With Unforgetter
What changes
The task bundle contains the 6 August decision, the reason it superseded the draft, the still-open capacity thread and your money-handling convention. The agent continues instead of re-deciding.
Decision
Fee measures against the original booking date
6 Aug · approved by you · supersedes the 22 July draft
Open thread
The capacity split is unresolved
Do not implement around it until it is decided
Method
Money is stored in minor units, never as a float
Your convention, carried into every agent that writes code
Scope
Booking product only
Finance and client material are outside this task
Provenance
3 sources, freshest 6 Aug
Every line traces back to a document you can open
The second agent opens with: "Continuing the cancellation fee. It measures against the original booking date, per the 6 August decision. The capacity split is still open, so I am leaving that untouched."
Scopes and data used
Booking product scope only
Decisions, open threads and coding conventions. Four months of transcripts stay out.
Expected effect
What you should notice
Fewer re-decided decisions and fewer late reverts. Context correctness and extra retrieval calls are measured against the same task.
StatusWorking now
Working reference task
The bundle shown is a sanitised scenario based on current product behaviour.
Agencies · project and company brain
Keep client scopes apart, by design rather than by discipline.
You run several clients through the same agents. The agents are excellent at reading everything, which is exactly the problem.
The job
Research a competitor for one client without touching another client's material.
Today
Where it breaks
Either the agent sees the whole workspace and you hope it stays in its lane, or you build a separate folder per client and keep them in sync by hand.
With Unforgetter
What changes
Each client is its own scope. The token used for this task can see the Northwind brief, the agreed method and the open questions, and can neither list nor retrieve other clients' material.
Decision
Northwind competes on service, not on price
2 Aug · agreed with the client · anchors every comparison
Method
Three independent sources minimum, each one cited
Your research standard, applied by whichever agent runs it
Open thread
Their pricing page is being rewritten
Treat anything quoted from it as provisional
Scope
Northwind workspace, 1 of 4 client scopes
The agent's token is configured for this project scope
Provenance
4 sources, freshest 25 Aug
Two client documents, two of your own notes
The second agent opens with: "Researching for Northwind only. I have their brief, the agreed method and the open questions. Nothing from your other clients is in this context."
Scopes and data used
One client scope of four
Client documents and your own notes for that account. Margin notes and other clients excluded and unlistable.
Expected effect
What you should notice
You can show a client exactly which material an agent could reach. Scope violations are measured with negative tests, and the target is zero.
StatusWorking now
Working scopes, prototype teams
Server-enforced project scopes work now. Multi-user roles and team isolation are in build and shown as a prototype on the Company Brain page.
Everyone · decisions and preferences
Corrections stick. Rejected ideas stay rejected.
The most expensive context loss is not a fact. It is a correction you already made, returning as a fresh suggestion.
The job
Stop every agent from proposing the fee rule you rejected on 22 July.
Today
Where it breaks
You explain the rejection in one tool. Next week another agent, or the same one after a memory reset, proposes it again with confidence. Your preference for minor-unit money handling has to be restated per project.
With Unforgetter
What changes
Rejected proposals are kept as rejected in the working truth and are excluded from task bundles as stale. Preferences are compiled as portable methods, so every coding agent applies the convention without being told.
Scopes and data used
Decisions, rejected proposals and preferences
Compiled with supersession dates. Protected policies cannot be changed by an agent.
Expected effect
What you should notice
Fewer repeated suggestions and fewer restated preferences. Proposal accept, correction and duplicate rates are tracked per agent.
StatusWorking now
Working now
Supersession and rejected-proposal handling run in the current consolidation flow. A visible review UI for proposals is in build.
Multi-agent · handoffs
A second agent continues from the first agent's state, not its transcript.
Handoffs fail because the receiving agent gets either nothing or everything. Neither is the working state.
The job
Let a research agent hand a finished brief to a coding agent, and let a coding agent hand a shipped change back to the record.
Today
Where it breaks
The handoff is a pasted summary. It is whatever the first agent thought was important, written for a human, with no scope and no provenance. The second agent asks the same clarifying questions again.
With Unforgetter
What changes
The first agent ends its session with a proposal: what was decided, what stayed open. The publisher reconciles it. The second agent's context request returns the reconciled state, scoped to its task, with provenance it can cite.
Scopes and data used
Shared project scope, separate agent identities
Each agent has its own token and trust zone. Proposals carry actor and evidence.
Expected effect
What you should notice
Handoffs that hold without a human relaying context, and false progress caught before it spreads. Both are outcomes the product is designed to make measurable.
StatusWorking now
Shared memory, no task relay
Unforgetter does not send tasks between agents. Agents share state through the memory; the handoff itself stays in your hands or your agents' own tools.
Not yet
What these pages do not claim.
- No team accounts, roles or shared billing. Small teams can join the alpha as a single workspace with one owner.
- No automatic capture from every client. Agents propose through MCP; some clients need an explicit instruction to do so.
- No published outcome metrics yet. Results are published only when they are measured and reproducible.
Private alpha
What should your agents remember?
Tell us which agents you use and, if useful, what you want their memory to keep track of. Applications are read personally.