Six AI agents run your company operations — engineering, research, growth, finance, ops — with SOP-driven execution, budget controls, and full auditability.
Strategy, delegation, sprint planning, code review
Implementation, debugging, PRs, test-driven development
Multi-modal R&D: codebase analysis, papers, market intelligence
Content creation, SEO strategy, analytics reporting
Monitoring, deployments, incident response
Revenue tracking, cost analysis, P&L reporting
Agents follow standardized operating procedures with evidence gates. Every step is verified before proceeding.
Safety enforced in hooks, not prompts. Regex-based tool blocking with role-based exemptions.
Per-role monthly budgets with automatic pause. No runaway spend. Usage-based billing via Stripe.
Row-level security, API key auth, scoped permissions, webhook delivery with HMAC signing.
Live task tracking, cost analytics, agent session replay, research corpus browser.
Auto-escalate model tier after 3 consecutive failures. Monotonic ratchet — agents adapt, not give up.
Tasks (Supabase) Dispatch Loop
┌─────────────┐ ┌─────────────────────┐
│ forge_tasks │────▸│ PrioritySemaphore │
│ forge_events │◂────│ Heartbeat + Scanner │
│ forge_costs │ └──────┬──────────────┘
└─────────────┘ │
▼
┌──────────────────────────┐
│ ForgeOrchestrator │
│ (task router + budget) │
└──────┬───────────────────┘
│
┌──────────┼──────────┐
▼ ▼ ▼
┌────────┐ ┌────────┐ ┌────────┐
│ CEO │ │Engineer│ │Research│ ...
│ (Opus) │ │ (Opus) │ │ (Opus) │
└───┬────┘ └───┬────┘ └───┬────┘
│ │ │
┌────┴──────────┴──────────┴────┐
│ Middleware Pipeline │
│ cost ▸ memory ▸ constraints │
│ audit ▸ evidence ▸ summarizer │
└───────────────────────────────┘
│
▼
Claude Agent SDK