AI AGENT ORCHESTRATION FOR SOFTWARE DELIVERY

Turn an idea into a tested pull request through agents you can see, control, and audit.

Coordinate research, planning, implementation, tests, and review in one visible workflow. Choose the model, tools, permissions, machine, and budget for every stage.

Create a free workspace Watch a real run

Bring your own provider credentials. Run Claude with Anthropic, Codex with OpenAI, or supported models through OpenRouter. Model usage is billed directly by your provider.

Simple Kanban board with research, planning, build, test, review, and pull request stages
Live workspace · hosted execution
Active modelCodex · buildStage override
Runtime$0.0818m elapsed
Evidence gate On trackChecks satisfied
  1. Idea
  2. Research
  3. Plan
  4. Build
  5. Test
  6. Review
  7. PR
01 / WorkspaceFree to configure
02 / RuntimePrepaid runtime
03 / ModelsModel usage billed by your provider
04 / CommitmentNo subscription

ONE MISSION, SPECIALISTS AT EVERY STAGE

Conduct the work, not just the prompt.

A card carries the mission through a policy-driven delivery line. Each stage can select its own agent behavior, model, tools, permissions, machine size, concurrency, and transition rules.

Pause, redirect, approve, or resume without losing the thread.

Simple Kanban is the execution and control layer for agents—not a replacement for your team’s project tracker.

CONTROL AT EVERY BOUNDARY

Give every agent a lane. Keep every run accountable.

01 / ROUTE

Route deliberately

Choose the model, skill, and reasoning level for each stage or card.

02 / CONTEXT

Keep context alive

Preserve session state, messages, documents, and decisions across handoffs.

03 / SCOPE

Limit the blast radius

Constrain tools, credentials, repositories, machines, and network access.

04 / ECONOMY

Know what it cost

See model route, runtime, machine, duration, and cost for every run.

THE SYSTEM AROUND THE MODEL

Capability needs guardrails.

A capable agent should get exactly the access, runtime, and budget its stage requires—then no more.

Allowed tools
Git · shell · scoped MCP
Read-only access
Secrets · production · billing
Runtime limit
45 minutes
Maximum cost
$0.20
Approval required
Before pull request
Repository scope
One branch · one repository

PROOF, NOT PROMISES

Every handoff leaves a readable trail.

Review the code path, tests, tools, decisions, model route, machine profile, elapsed time, and runtime charge before the workflow advances.

  • Diff
  • Tests
  • Tool calls
  • Decisions
  • Model route
  • Machine
  • Time
  • Cost
  • Approvals
01 · WorkflowStages, agents, models, and runtime policies in one board.
02 · Run historyModel route, machine, duration, output, and charged runtime.
03 · SessionContinue the same agent conversation without losing context.

PAY FOR EXECUTION, NOT ANOTHER SEAT

Start free. Fund the runtime when the workflow is ready.

Your workspace is free. Credits fund hosted agent runtime. Model usage is billed separately by your provider.

Before the runSee a maximum estimate from the selected machine, runtime limit, and per-run minimum.

After the runSee final elapsed time and runtime charge in run history.

Maximum estimates assume an effective Fly performance machine and are rounded up to the next cent per execution, with a $0.05 minimum. Shared CPU runs can cost less. You see the final charge in run history.