Aigon Multi-agent coding harness
Write a spec once. Launch competing implementations. Let one agent judge another. Merge the diff that survives review. Do it all from an intuitive kanban board.
Aigon ships four modes — Drive, Fleet, Autopilot, and Swarm. Here's Fleet in 30 seconds:
Claude Code and Cursor compete, Codex evaluates — best implementation wins.
1. Fleet started — agents implementing in parallel
2. Both submitted — Codex evaluating
3. Evaluation complete — winner, Claude Code merged, with the best from Cursor implementation
Why use Aigon?
01
Different model families catch different classes of bug. When Claude reviews Claude, the reviewer shares the implementer's blind spots. Aigon Fleet runs Claude Code, Codex, Cursor CLI (and more) in parallel against the same spec, with a different coding agent as the merge reviewer.
02
Agent subscriptions hit 5-hour quota limits. Aigon polls your quotas every 30 minutes, surfaces headroom on the dashboard, and auto-fails over to whichever coding agent still has quota. Three subscriptions, one elastic pool. Schedule specs to run when you have quota or while you sleep -- no more FONUT (Fear Of Not Using Tokens).
03
Aigon never sits in your billing path. Spend the Claude Pro, ChatGPT Plus, and Cursor Pro subscriptions you already have — Aigon orchestrates the agent CLIs, not their underlying model APIs. No tokens, no markup, no metering.
04
Each agent runs in its own git worktree under tmux — tmux attach to any session and watch the work happen live. Specs live in your repo as Markdown; every workflow move is a git diff. No cloud sandbox, no opaque execution.
On disk, in your repo
Each feature is a Markdown spec in your repo — not session state, not chat history. Build a feature, ship it, then run the same spec through a different agent if you want a fresh take. The spec is the constant; implementations come and go.
Aigon doesn’t just store specs — it links them into a graph. Research findings feed feature specs. Feature implementations feed evaluations. Evaluations feed the next round. By the tenth feature, the eleventh agent doesn’t start cold — it reads the source code AND the full chain of intent that led to it.
Most coding tools only handle “write the code.” Aigon ships three workflows that link: research investigates open questions, features deliver shipped code, feedback triages user input. Research findings become feature specs. User feedback becomes prioritised work. The loop before and after the build, not just the build.
CLI in action
How it works
Choose your mode
Hands-on + one agent
Use when you want tight control over implementation details and review checkpoints.
Aigon
aigon feature-start 07Outcome: one guided implementation branch with a full spec and log trail.
Hands-on + multi-agent
Orchestrate competing implementations you can evaluate and adopt the best from.
aigon feature-start 07 cc gg cxOutcome: parallel worktrees and comparable outputs for structured selection.
Hands-off + one agent
Use when the scope is clear, but you want the peace of mind of a reviewing agent.
aigon feature-autonomous-start 07 cx --review-agent=cc --stop-after=closedOutcome: automated implement-validate loop that stops when checks pass or budget ends.
Hands-off + multi-agent
Fully orchestrated, fully autonomous — parallel agent runs converge into comparable outputs ready for auto evaluation and completion.
aigon feature-autonomous-start 07 gg cx --eval-agent=cc --stop-after=closedOutcome: concurrent autonomous runs across agents with logs ready for comparison.
The steps
Create a feature spec and move it into the backlog.
Your workflow, visualised
The Aigon Dashboard is the visual way into your spec-driven workflow. Same pipeline, same agents — but managed through a browser UI instead of CLI commands. Drag features across columns, launch agent sessions with one click, and watch your pipeline move in real time. No terminal required.
Visual Workflow
Move features through your development pipeline with a familiar Kanban interface. Every column maps to an Aigon workflow state — no commands to memorise.
Monitor
The monitor tab shows running sessions, attention items, and recent events across all your repositories — everything you need to know at a glance.
Reports
Five aligned charts — features completed, commits, cycle time, commits per feature, and rework ratio — show whether your team is shipping faster, writing tighter code, and fixing less. Summary cards, agent leaderboard, and filterable detail tables give you the full picture.
Telemetry
Track token usage and cost across Claude, Gemini, and Codex sessions — broken down by activity type (implement, evaluate, review) and attributed per agent. See exactly where your spend goes. Learn more →
Remote Access
The dashboard runs on your local network — open it on your phone or tablet over LAN, or use Tailscale to check on your agents from anywhere. Monitor from the couch.
Workflow Intelligence
Five aligned charts track the metrics that matter: are features shipping faster? Are agents producing cleaner code? Is rework going down? Reports gives you the data to answer these questions — across every repo, every agent, every period.
Features completed and commits over time — stacked to show feature-attributed vs non-feature work.
Median cycle time and commits-per-feature trending over time. See whether features are getting tighter.
Percentage of fix commits per period. Trending down means agents are landing correct code on the first pass.
Get started
npm install -g @senlabsai/aigon@nextInstall Aigon and run your first feature loop in under five minutes.
Three workflows, four execution modes, and how Aigon organises your work.
Step-by-step walkthroughs for Drive, Fleet, Research, and Dashboard.
Every Aigon CLI command, organised by feature, research, feedback, and infrastructure.
CLI commands, configuration, agents, and file structure.
Browse the full docs site when you want the whole map in one place.
Community
Contribute specs, improve workflows, and share real-world patterns for running multi-agent engineering teams effectively.
Aigon Pro
Aigon Pro is the layer between “I shipped a feature” and “my workflow ships features” — autonomous runs on a schedule, agent quality metrics with AI coaching, benchmarks, and portable state across machines.
