Largami

Design Orchestrate Evolve

Teams of AI agents, designed and directed by you.

Largami is a visual workspace for AI-assisted work. Describe what needs doing, shape the team of agents that will do it, approve the moments that matter, and see what was done — and what still needs proof.

See how it works In private development
Largami's Design view. A workflow called Graceful shutdown: a goal, then an App Engineer and a Platform Engineer working together, a Shutdown Reviewer, an Approve rollout decision, and a Rollout Engineer. Beside it, the execution flow lists the same steps in order, with 4 agents, 5 requirements and Ready to run.
The Design view of a real workflow: two engineers work in parallel, a reviewer checks them, you approve the rollout, and only then does it run.

How it works

From a plain sentence to work you can stand behind.

Each project keeps its conversations, workflows and runs together. Here is one piece of work, end to end: a fix for an OpenShift service whose pods took 30 seconds to shut down.

  1. 01Describe

    Say what you need, in your own words.

    Start with a conversation. The AI proposes a team: who does what, in which order, and why. It arrives as a draft you can read, edit or dismiss. Nothing changes until you apply it, and nothing runs until you authorize it.

    A conversation with a proposal ready for review: the proposed connections between agents, the reasons for the design, and the buttons Apply to draft, Edit and Dismiss, above the note that nothing runs until you authorize a run. Beside it, the proposed team as a graph.
    A proposed team, with its reasoning. Applying it changes the draft only, as one step you can undo.

  2. 02Design

    Shape the team on a canvas.

    Each agent has its own instructions, model, effort, tools and folder access. Connect agents so one passes its result on, assigns work, or sends work back for another round. Add an agent by describing it, and put your own decisions where they belong.

    The Design view with a proposed change: a new Rollout Verifier after the Rollout Engineer, marked Changed. The inspector shows the proposal with Apply, Refine and Discard, and the request that produced it: set up this new agent to verify the rollout under live traffic.
    A verifier added by describing it. The change shows in the graph and waits for Apply, Refine or Discard.

  3. 03Run

    Run it, with you at the gates.

    Preview exactly what will run (agents, models, prompts and limits), then authorize it. Agents work in parallel where they can, each in its own sandbox, and you can follow every one of them as it works. The run stops where you placed a decision, and commands that need permission ask you first.

    A run waiting for you. The two engineers and the reviewer are done, Approve Rollout needs you, and the Rollout Engineer waits for it. The approval panel shows the reviewer's notes and three choices: Approve, Request changes with a note, or Reject.
    A real run at its approval gate: approve, request changes with a note, or reject.

  4. 04Review

    Read results that don’t overstate.

    A finished run shows what was made, what it cost, and which requirements were proven. Requirements are checked by commands or judged by reviewers, and anything that couldn’t be established is reported as exactly that.

    The Results of the finished run. Headline: Finished, but 3 requirements could not be established. A Needs attention list shows three requirements judged unsure, and the files the run made are listed with the agent that made each one.
    The same run, finished: the work is done, and three requirements are honestly marked “unsure”.

  5. 05Evolve

    Change course without starting over.

    Requirements change. Edit a step or a requirement on a finished run, and Largami works out which work still holds, what has to run again and which decisions come back to you, before anything starts.

    Evolve on the finished run. The engineers, the reviewer and the Rollout Engineer are marked Reuse, each with the cost of the work kept; Approve Rollout asks you again; the Rollout Verifier is marked Run, changed. Beside it, the request: also report p95 latency during the rollout.
    A new requirement on a finished run: four steps’ work is reused, the changed verifier runs again, and the approval comes back to you.

Principles

Built for oversight, not just automation.

  • Nothing runs without you

    Plans are previewed before they run. You authorize them, decide at the approval points, and allow or deny the commands that need it.

  • Everything on the record

    Each agent’s instructions, model, inputs, activity, files and cost are kept and visible: live while it works, and afterwards.

  • Your folder stays untouched

    Each run works on its own copy of the project. Nothing is copied back unless you choose to.

  • Every change can be undone

    Designs keep their history: undo a change, save revisions to return to, and see exactly what each run was made from.

Status

In private development.

Largami is being built and used privately. It isn’t available to download or buy yet, and it isn’t open source. Today it runs locally, on the computer it is installed on, and its agents run on Claude Code.

If you’d like to hear more, or follow its progress, write a line.

contact@aminelg.com