Client Dossier
Who decides, how they like to be written to, hard boundaries and the history of the relationship.
changesweeks to months
Altnull
What if your whole team used AI like your best AI user? Each draft starts with who the client is, their history, and what you're working on for them right now.
Built for B2B service firms of 5 to 50 people. A person reviews every draft before anything is sent.
Can we still hit the 12th? Also, did the scope change affect the estimate?
Example with invented names. Facts shown here follow the format Altnull stores in a firm's Drive.
The problem
Your clients trust you because you remember them. Your team can too.
Staff answer from part of a thread. The draft comes out polite and generic, and it misses the current phase, the open blocker and the way this client likes to be written to.
Handoffs drop commitments because nobody logged the note. What you know about a client lives in inboxes and in people's heads.
Several accounts end up in one chat, and one client's details turn up in another client's reply. You worry about that every time someone pastes a thread into an AI tool.
How it works
Altnull reads your sent and received email and keeps what it learns as plain files in your own Drive.
It writes down who the client is, their history and where each project stands. Anything important, such as a new deadline or a scope change, reaches you as a one-click confirmation.
Open a client email and a draft is already in the thread. You review it, change it and send it.
What you can count on
The detail
The next sections show what is stored, how facts are sorted by risk, how clients are kept apart and what we hold.
The Context Triad
Each lives in its own Markdown file because each changes at a different speed. Anyone can open them, edit them and take them if they leave.
Who decides, how they like to be written to, hard boundaries and the history of the relationship.
Phase, owner, milestones, blockers, next action, key dates and the decision log, with a client-visible summary.
Your SOPs, scope rules, escalation paths and approved templates, written by firm leadership.
Each entry records its source, capture date, confidence and whether a person confirmed it. Facts with a deadline are flagged for re-check after a set time, so stale blockers do not quietly steer a draft.
When a person edits a file, that edit wins over any later automatic change. If the two disagree, Altnull shows the conflict and does not overwrite.
## Phase 2 launch - Target date: 2026-11-12 - source: email from J. Rivera, 2026-10-02 - confidence: high | confirmed by: A. Chen - review by: 2026-11-01 - Blocker: awaiting client sign-off on scope change - source: call transcript 2026-10-01, 14:32 - confidence: medium | confirmed: no
Ingestion
Altnull reads sent and received email, sorts every candidate fact by risk and confidence, and writes without a person only in the safest tier. Impactful changes arrive as a short digest of one-click confirmation cards.
| Tier | Examples | What happens |
|---|---|---|
| Low risk | New contact and title from a signature, a meeting that took place, a file that was sent | Written with provenance and listed in a daily change log |
| Impactful | Deadline change, scope change, client approval, pricing mention | One-click confirmation card to the thread owner |
| Tentative | "Let's aim for Nov 12", hypotheticals, forwarded text with an unclear sender | Not stored as fact |
| Out of bounds | Legal terms, payment details, threads that cannot be tied to one client | Never written automatically |
Client isolation
Firms do legitimately run one analysis across many clients, so Altnull enforces boundaries by mode and at the point of output instead of banning multi-client work.
| Mode | Context loaded | Allowed output |
|---|---|---|
| Client | One client's dossier, portfolio and the playbook | Client-facing drafts, reports and emails. Nothing from any other client enters. |
| Cohort | Clients you choose, such as ten in one industry, if their settings allow it | Internal work only, labeled as internal. Client-facing output means switching to one client. |
| General | Playbook and firm knowledge, no client data | Internal work and drafting from scratch |
Every stored item carries a client ID. Before a draft is sent or exported to one client, an output check blocks it or asks for a scrub if anything in it traces to another. Each client is marked usable as an example, anonymized only, or never.
Retrieval permissions and the output check are the hard guarantees. The model's own behavior is an extra layer. Text pasted into a chat has no client ID, so the mode banner and staff guidance cover that case, and release tests include pasted content.
Data handling
Roadmap
The same context serves three surfaces, built in this order. Each stage starts when the one before it clears a measured gate.
Detects a client email, loads that client's dossier and project state, and creates an in-thread Gmail draft. Staff review, refine and send.
An MCP server gives Claude and other chat tools client-scoped context. Chats can propose updates, and a person confirms them. Support is verified tool by tool.
Project-type playbooks, then call transcripts from the recording tool your firm already uses.
A shared workspace portal with time tracking, and a SharePoint adapter.
Design partners
We are recruiting three pilot firms for the first 90 days. Every pilot produces numbers: draft acceptance, facts that survive into the sent email, extraction accuracy and minutes saved per reply.
Altnull. Know your client. A little more with every email.
hello@altnull.com
A good fit