ARCHITECT'S GRID Sign in

powered by NTH Iteration1.0.0.0 · In development

Private beta · local-first

Run an AI organization.
You hold the chair.

Architect's Grid appoints a team of AI agents around your goal, to build something, to run something, or both. They deliberate in isolation, raise a hand when they have something to add, and propose. Nothing becomes a decision until you approve it.

Join the waitlist See the three missions

Charted in your browser from a template. Nothing is sent.

Or watch
01 — The protocol

A council that speaks when you give it the floor.

Not a chat with many voices: a chamber with rules. Who is appointed, what each agent may see, who speaks next and what counts as decided are yours to set, and the server enforces them. Every step leaves a record.

  1. Appoint

    appointment v3 · mandate · model qwen3.5:4b · 4 pinned sources

    Give each agent a mandate, a model and the sources it may read. Appointments are versioned; changing one invalidates work that no longer fits.

  2. Open a topic

    topic · 7 agents · 7 private contexts · 0 shared drafts

    Ask the question once. Each agent reviews it privately, in its own context, and never sees another agent's drafts.

  3. Hands, not noise

    hand · netcode · important · “Co-op cannot hold four players yet.”

    Agents with something worth saying raise a hand with a one-line summary and a priority. Silence is an answer too.

  4. Grant the floor

    floor → netcode · streamed · shared on completion

    You choose who speaks. One speaker at a time, streamed live, shared with the others only when the speech is complete.

  5. Decide

    decision v4 · approved by the chair · evidence: 3 sources

    Proposals stay proposals. Rules, roadmap changes and commits become canonical only on your approval, with their evidence attached.

02 — Three missions

Build it. Run it. Or both.

The same grid organizes very different work. Pick the mission; the organization, its teams and its rhythm follow. Scroll to watch each one take shape.

Model cost▮▮▯▯▯Pace▮▯▯
01 · Build

Build a game. Then a whole online world.

Start with an indie game: a director, level designers, a netcode engineer and playtesters. Grow it into a multiplayer studio, then an open-world online game: divisions of teams, each with its own leads. Specialists plan, build and check; nothing lands without you.

What your grid does

  • Turns your idea into a plan. Onboarding drafts a charter, a dossier and a roadmap; a critic reviews it; you approve what becomes the plan.
  • Staffs the studio. Design, engineering and playtest teams from an organization template, each with a lead you talk to.
  • Writes code as proposals. A Builder agent drafts change sets, a reviewer checks them, anyone can object.
  • Tests safely. Runs happen in a Docker sandbox with no network, no privileges and hard limits.
  • Keeps the record. Commits happen on your approval; project docs are generated from the approved record.

You decide the roadmap, every appointment and every commit.

02 · Build + Operate

Build the desk, then run it every day.

From a market-analysis desk to a multi-asset research desk to a quant research platform: a data team builds and maintains what the others depend on, research makes the recurring calls, risk and review challenge both. Analysis and review only: no orders are executed.

What your grid does

  • Separates responsibilities. A data team builds the pipeline, an assessment team uses it, a review team questions both.
  • Connects to the real world. REST and WebSocket APIs, Prometheus and read-only SQL through one gateway; every agent write waits for your approval.
  • Keeps interfaces honest. APIs are captured as versioned contracts the agents share, with drift warnings.
  • Keeps work apart. Work orders claim files and resources, so two teams never edit the same thing at once.
  • Makes decisions reviewable. Decision packages carry evidence and sign-offs before anything becomes canonical.

You decide which connections exist, who may write, and what becomes a rule.

03 · Operate

Run a race weekend. Then a whole season.

No software to ship: a race to run. Strategy, telemetry and the pit wall argue over the rain and leave the call to you. Grow it to a two-car team, then a championship season with a factory, logistics across continents and a cost cap to respect.

What your grid does

  • Coordinates specialists around one shared plan, shared records and your constraints.
  • Meets when it matters. You schedule the meetings; agents ask to meet through the organization, and you approve.
  • Speaks only with a reason. An agent raises a hand when it has something to add; you give the floor.
  • Follows through. Commitments and follow-ups carry on between meetings.
  • Rests without forgetting. Pause or archive the organization; its decisions and memory stay.

You decide the goal, the plan and every change of course.

Examples, not templates. A role describes a remit, not verified expertise: agents advise and prepare, they do not replace qualified guides, physicians or licensed professionals. Data sources and integrations are configured per project, and each use needs its own validation.

03 — A replay, not an animation

Three weeks of a project, replayed from its journal.

This is the map on Project Home, drawn from a demo project's roadmap, organization and recorded status changes. Where a hill flattens, work was accepted; where a storm stands, work failed or waits on a prerequisite; teams stand where their work was in progress. Drag the history, or press Replay.

Demo data: a synthetic project built with tools/demo_showcase.py through the app's own API; its history is backdated. Nothing here is a customer's work.

04 — Capabilities

What your grid can do for you.

Everything below exists in the app today. Everything that acts outside the grid does so through permissions you grant.

A council chamber

Topics, isolated checks, raised hands with priorities, one speaker at a time, breakouts for side questions, and a record of every turn.

An organization

Teams and leads from templates, a leadership council you talk to, and meetings that follow the chart.

Code with a safety net

Change sets, reviews and objections; tests in a network-less sandbox; commits only with your approval.

Tools and APIs

Named connections with per-agent access; secrets never reach an agent; writes wait for your approval.

Local first, budgets always

Runs on your machine by default. Cloud models only when you turn them on, within a budget you set.

Everything in one place, provenance kept

Browse your whole grid: repositories, each agent's context and work, logs and your own files. What the system uses stays read-only; decisions keep their sources and versions.

05 — Guarantees

Built so authority never drifts.

The guarantees live in the server, not in a prompt. They hold whichever model you run.

Human authority

Only you appoint, approve and delegate. No model endpoint can grant itself the floor, a budget or a decision.

Local first

New workspaces run on your own machine through Ollama. An API key on file never switches a project to the cloud by itself.

Sealed by default

Every account runs in its own process with its own data. Projects read each other only through links you grant.

Provenance kept

Decisions keep their sources, versions and the exact appointment that produced them. Excerpts carry their warnings.

Private beta

Registration is closed. The grid is not.

We are onboarding a small number of organizations at a time. Tell us what you would build or run, and we will reach out when a seat opens.

Join the waitlistSign in