Flowmanner Open a workspace

AI workflow platform

Every mission,
on the record.

Flowmanner plans a mission as a graph you can read before it runs, executes each node on your own host, and writes an audit entry for every memory it keeps. Nothing about your workspace is inferred after the fact.

Open a workspace Read the record

Existing keys work. A registered local model costs $0 per token.

Fig. 01 One mission, five nodes — plan, tools, verify, ship, audit. running
Method

Plan in public. Execute in private.

The graph is the contract. What ran is what the graph said would run.

  1. i

    Build the mission

    Drag nodes onto a canvas: retrieve, call a tool, branch, verify. The blueprint is a document, reviewable before a token is spent.

  2. ii

    Run it on your terms

    Each node executes in an isolated workspace with its own volume. Stop at a node, ask for permission, or send the turn down a branch.

  3. iii

    Accept the record

    Costs attributed per agent, memory writes logged with before and after. The mission closes with evidence, not with a green check.

The record

Four things we keep, and why.

This is the part most platforms leave to a log file and a promise.

Workspace volume
One sealed environment per workspace: its own files, indexes and configuration. Separation is a property of the deployment, not a filter in code.
Memory timeline
Every store carries actor, timestamp and before/after. Pause it, forget a fact, correct it — each action lands in the same timeline.
Per-user retrieval
Context is assembled by a query scoped to user_id at the database level. Another user's document cannot be reached by application logic.
Cost attribution
Token spend is booked per agent and per mission, and the self-hosted path books zero. Sensitive work can run where the meter does not move.

Bring the keys you already have.

Open a workspace, point it at a model, and run one mission end to end. The record starts at the first node.

Open a workspace
Limits

What this is not.