A local-first desktop app that turns a plain-English description of your business into a working org of AI agents — a CEO that delegates to department heads and workers, and hands back real reports, plans and spreadsheets. On your own machine.
One instruction, delegated to three departments and a worker, synthesised back into a single deliverable.
Watch a single instruction travel down the org and come back as a finished piece of work.
Recorded on a demo org with synthetic data. Any figures on screen are illustrative.
What it shows. The CEO agent reads context, then delegates to the Marketing, Sales and Finance department heads — Sales in turn pulls in its own analyst — and synthesises their work into one consolidated go-to-market plan, with a section per department. The delegation streams live as it happens.
A loop around a single model is one improviser doing everything in prose. SIGMORA runs a bounded tree of specialised agents — each with its own instructions, its own model, and only the tools it was granted.
In plain English. It generates a complete agent org — the roles, the delegation tree, starter workflows and a data plan.
Later, an “Improve my org” advisor reads your real run history and proposes concrete changes.
A tree of agent nodes: a CEO leading department heads for Marketing, Sales, Finance and the rest, which in turn lead workers.
Each node has its own role, instructions, model, tool allowlist and direct reports. Nothing is shared implicitly.
Every manager holds a delegate tool scoped to its direct reports
only. Each child runs fresh and isolated; results bubble up and the lead
synthesises the answer.
Anything carrying a number is assembled by code, not the model — totals, metrics and tables are computed exactly.
Autonomy you can actually leave running.
Hard caps on delegation depth, agent runs, iterations, per-tool call budgets and wall-clock time. The clock is shared across the whole run, so one limit winds the entire tree down.
Any risky or irreversible tool is wrapped in a human approval gate. It pauses and waits for you before it runs.
A full audit trail of every start, every delegation, every tool call and every result.
Most tools ask you to trust a policy. This is a desktop app, so what leaves your machine has an answer you can enforce — not a paragraph you have to believe.
The lead and every worker under it move onto a local model, and all cloud traffic is blocked. Not a default that a per-agent override can quietly undo.
Cloud models are bring-your-own-key and stay inert while it is on. Local models run through Ollama — no key, no account.
A packaged desktop app on your own hardware, alongside a web UI and a terminal client for the same engine.
Your knowledge base is a local Markdown vault — ordinary files in ordinary folders that your agents read from and write to.
Connectors are the one surface that can reach outside, so they are what Private mode cuts off. Each is MCP-backed: you register the server, choose which tools it grants, and attach it to the agents that should hold it. This list is generated from the product's own connector catalogue, so the site cannot claim an integration the app does not offer.
Departments, workflows and automations an agent org can run.
The Finance Agent matches invoices to what was actually delivered, flags the gaps, and shows its working — every figure traced to a source row.
Explore →Operations agents watch the queue, escalate what's stuck, and post a single shift summary to Slack instead of a stream of separate pings.
Explore →The Research Agent reads your vault, the account's recent mail and the notes in Sheets, then returns a one-page brief on what changed and where you fit.
Explore →There is no installer to download, so the honest version of “try it” is this: a small number of businesses run SIGMORA on their real workflows, and what they run into decides what gets built next.
Who it suits. Businesses with repetitive back-office work — finance ops, vendor management, reporting — where the same job comes round every week and somebody senior is still doing it by hand.