Aigon: A Spec-Driven Development Framework for Multi-Agent Collaboration

Aigon is an open-source multi-agent orchestration system by Sen Labs that enables collaboration among mainstream AI programming assistants like Claude, Gemini, Codex, and Cursor. Its core approach uses Markdown specifications to drive development processes, allowing multi-model competition and optimal result merging. It aims to combine AI programming efficiency with engineering controllability and auditability.

With the rapid popularity of AI programming assistants (Claude Code, Gemini CLI, Codex CLI, Cursor), developers face the challenge of getting these tools to work together instead of in isolation. Aigon addresses this gap by providing a structured way to orchestrate multiple AI agents, bridging individual tool use and coordinated, engineering-driven development.

1. Spec-Driven: All work starts with a Markdown specification stored in docs/specs/ (version-controlled with code), ensuring auditability, traceability, human-AI collaboration, and no vendor lock-in.
2. Multi-Agent Orchestration: Fleet mode allows multiple agents (e.g., Claude and Codex) to work on the same spec in parallel, leveraging each model's strengths (e.g., Claude for architecture, Codex for code details).
3. Vendor Independence: Supports all major AI programming assistants (Claude Code, Gemini CLI, Codex CLI, Cursor) with flexible agent selection.

• Feature Workflow: Inbox (quick idea recording) → Backlog (detailed spec with user stories) → In Progress (Drive mode for single agent, Fleet mode for multi-agent) → In Evaluation (compare results) → Done (merge optimal implementation).
• Research Workflow: Parallel research by multiple agents on different tech directions, then synthesize findings.
• Feedback Workflow: Collect user feedback as Markdown docs, deduplicate/classify, and promote valuable feedback to specs.

1. Slash Commands: Directly in AI agents (e.g., /aigon:feature-now dark-mode ...) to create specs, assign IDs, and start development without leaving the agent interface.
2. CLI: For terminal users (e.g., aigon feature-create, aigon feature-start for Drive/Fleet modes, aigon feature-eval).
3. Web Dashboard: Localhost-based dashboard showing feature/research/feedback status, real-time agent sessions, commit activity, telemetry charts, and full project index.

Suitable Scenarios: Important feature development (Fleet mode for quality), technical research (parallel solutions), code refactoring (multi-angle refactoring), team collaboration (specs as common language), open-source maintenance (systematic feedback handling). Unique Value: Quality assurance via multi-agent competition, traceable processes, reduced AI hallucination risk (human review in spec phase), long-term knowledge accumulation (spec docs), developer control over final decisions. Comparison: vs traditional AI assistants (adds multi-agent, spec-driven, built-in evaluation); vs commercial platforms (data privacy, no subscription, full control, customizability).

• Aigon Pro: Commercial version planned with AutoConductor (unattended automation), deep insights (agent quality trends, cost analysis), AI coach (team optimization suggestions). Open-source core remains free.
• Community: Maintained by Sen Labs (Apache 2.0 license), active on GitHub (docs at aigon.build/docs, discussions, issues, contribution guide). Encourages community contributions for bug fixes, features, and docs.