OrgMapAI
The problem
"AI assistant" sidebars in CRMs, copilots in spreadsheets, summarizers on top of email. Each one is an island. None of them know the org around them.
Forwarding, summarizing, filing, chasing follow-ups. The "between-the-tools" tax. Small teams spend more time routing work than doing it.
Six tools, ten dashboards, no shared memory. By the time it ships, the workflow has changed.
OrgMapAI is built ground-up around your organization — not strapped onto the tools.
The shift
Stop wiring AI into your tools. Wire it into how you already delegate work.
Old model
- Pick a tool (Slack, email, CRM).
- Bolt an "AI assistant" onto it.
- The assistant knows that tool's data — nothing else.
- You manually ferry context across tools.
OrgMapAI
- Define your org chart.
- Each role is a seat — fill with a person, an agent, or both.
- Drop work on the seat: it executes, delegates down, or escalates up.
- Memory + integrations live on the role, not the tool.
People already understand this mental model. We just give the org chart a brain.
How work flows
Example: an inbound RFP from a prospective client. Issues become tasks, tasks fan out into delegations, delegations return outputs, and outputs roll back up for review.
Every step is auditable: who decided what, with which inputs, costing how many tokens (if AI was used).
The org chart, made of seats
Sample: Acme Coffee Co — a small DTC roaster. Each role is a seat: fill with a person, an agent, or both. Same role library across every company; reporting lines drive how work routes.
Drag-to-reparent in Tree mode. Rename, reassign, or split a seat across companies in seconds — the chart is the source of truth.
Each seat does work — your way
A person fills the seat. AI auto-reply is off. The role still has chat threads, tasks, outputs — but the human drives every reply. Use this when you're hiring slowly or want pure manual control.
A person fills the seat. AI assist is on. Tasks land in their inbox; the agent drafts; the human reviews and approves. Best of both — speed of AI, judgment of a person.
No human assigned (yet). The agent runs autonomously inside your guardrails — write tools require an explicit toggle, and tasks default to "review required."
Bring any model
Add API keys for whatever model family you want — Anthropic, OpenAI, Google, xAI, DeepSeek, local. Set per-role, fall back per-workspace. Each inference is metered to the right key.
Connect anything
Live today: Gmail, Calendar, Drive, Telegram. Roadmap is open: any OAuth provider or API-key service can plug in. Tool-grants per role, with scope subsets and revocation.
Chat is everywhere
Every node, role, issue, and task has its own thread. Drop a file, @-mention anyone, the right seat picks it up — and follows through.
Drop a file. Get an action.
Anything you can paste into chat becomes live context for the seat that picks it up.
📷 Images
Schedules, whiteboards, screenshots, receipts, signage, even a phone photo of a sticky note.
Sent to the model as a native vision block. The agent reads the image directly — no OCR pre-processing, no manual transcription.
📄 PDFs
RFPs, contracts, leases, invoices, multi-page reports, scanned forms.
Native PDF support: every page including layouts, tables, and embedded images flows into the agent's context.
📎 Other files
CSVs, spreadsheets, JSON exports, plain-text docs, code, transcripts.
The agent gets the file as a linked reference and a description of what's inside; can read it via the file-read tool when needed.
🔒 Workspace-scoped storage
Every upload lives in private, encrypted storage scoped to your workspace.
RLS gates every read/write. Signed URLs with 30-min TTL when an agent fetches a file. Revoke a member, the access goes with them.
Coming next: drag-and-drop into any thread, paste-from-clipboard, and inline preview before send.
Integrations: granular, revocable
No "all-or-nothing" OAuth grants. Connect at the workspace, grant per seat, revoke per seat.
Connect once
Workspace owner adds Gmail / Drive / your custom API once. Credentials live in an encrypted vault — never in app code, never in logs.
Grant per seat
CEO gets full Calendar. Personal Assistant gets events.read + events.write, nothing else. Revoke either independently.
The integration framework treats every provider the same — OAuth flow, scope subset, revocable grant, audit log. New providers ship in days, not quarters.
Bring your own keys, your own models
No model lock-in, no markup, no surprise bills.
🔑 BYOK
Paste API keys once at the workspace. Every agent run bills to the right key automatically. Keys are encrypted at rest and never appear in app code or logs.
↓ Workspace fallback
Agent-only roles bill to the workspace owner's key by default. Invited collaborators never accidentally burn your quota — they're billed to theirs.
📊 Per-token metering
Every inference and tool call is logged: model, prompt + completion tokens, dollar cost, billed-to key. Pivot by role, company, or time range.
🎯 Right model per seat
Big-context for the CEO. Fast & cheap for the Personal Assistant. Code-specialized for Engineering. Set per-role; switch any time.
Multi-company portfolios
Run a holding company, an agency with sub-brands, a family office, or personal + work on one login.
↕ Move companies
Reparent a subsidiary in one click. Cycle-safe; descendant ltree paths cascade automatically. The Acme Wholesale node above can move under any other parent without losing its tree.
⊕ Split allocation
One person can be CEO at three companies. Set 50/30/20 splits or hit Split evenly. Warning if you over-allocate.
👁 Subtree visibility
At Portfolio HQ, see every issue / task / output from every subsidiary rolled up into one feed. Drill into any company for its scoped view.
🔒 Subsidiary autonomy
Each company has its own integrations, billing key, RLS scope. Move someone in or out without losing their work. Sell a subsidiary? Detach it cleanly.
What this looks like in practice
Five very different orgs, same primitives.
Drop a kids' sports schedule in Personal Assistant chat → calendar events created, conflicts surfaced. Forward an offer letter PDF → benefits comparison drafted, follow-up questions for HR queued in your Inbox.
RFP arrives → CEO files an issue, dispatches "Cost estimate" to Estimator and "Brand narrative" to Marketing Lead. Each agent drafts in parallel; VP Product reviews and merges into a final proposal with one click.
Client uploads markups → Architect reads, files three issues for Interior Designer + MEP + Structural, dispatches each. The Project Manager (human) reviews the rolled-up output before sending back.
Founder is the CEO + on-camera talent. Personal Assistant agent triages email and DMs. Content Marketing role (human + AI) drafts captions; founder approves. Scheduling agent books podcasts and confirms the night before.
Treasurer (human + AI) drafts monthly financials from uploaded bank statements. Secretary agent turns recorded meetings into minutes the next morning. Maintenance coordinator (human) triages owner complaints; an Inbox role agent batches non-urgent ones into a weekly digest.
Get started in five steps
From sign-in to first agent acting on your behalf — about ten minutes.
Your personal workspace is created automatically. Free tier; no card required.
AEC, SaaS, Agency, Family, etc. — or start blank. The template instantiates a full team of pre-prompted roles in one click.
Invite a person, leave it as an agent, or do both. Flip Allow AI to auto-reply per role to control human-vs-agent voice.
Gmail / Calendar / Drive / Telegram in seconds; OAuth or paste-a-key for the rest. Grant per role with scope subsets.
Drop a file in chat. File an issue. Dispatch a task. Watch it route through your org and come back as a reviewable output.
Built on Supabase + Cloudflare + your model of choice. Open architecture. You own your data.