Why it exists
If you tried to build with agents in 2024 or 2025, you probably hit the same walls we did. You ended up with twenty terminal windows open, one per agent, and no way to see what was running where. You had no budget, no caps, and you watched a looping agent spend hundreds of dollars in a day with nothing to show for it. You had no org chart, so agents could not delegate to each other. You had no audit trail, so when something shipped broken you could not reconstruct what happened. You had no memory, so your hundredth run was the same quality as your first. Company Agents is the layer that fixes all of those. Keep your Claude Code. Keep your Codex. Keep your Gemini CLI. Company Agents gives them a company to work inside.What you get
A real org chart. Humans and AI agents share the same structure. Every agent has a name, a role, a reporting line, and a manager who approves the work. Every decision lands in an audit log, signed and timestamped. Stacked budgets. Spend caps at six scopes: company, team, agent, workflow, task, and loop. Every inner cap protects the outer one. No agent can ever spend more than its team. No team can ever spend more than the company. Overruns escalate to a human instead of silently drawing from the next scope up. Workflows with gates. Multi-stage pipelines where each stage is assigned to an agent, and every handoff goes through a gate. Fast programmatic checks run first, LLM reviews second, human approval for anything risky. Memory that compounds. Every agent has notes at four scopes: per-agent, per-project, per-client, per-company. Lessons that prove themselves at a narrow scope automatically promote to broader ones. Your hundredth website build is nothing like your first. An Executive Assistant in your sidebar. A dedicated AI that works across every company you run. Configure agents, install integrations, analyze costs, delegate work, all by asking in plain English. Twelve system-level tools plus every tool your agents have. Bring your own agent runtime. Eight CLI adapters at launch: Claude Code, Codex, Gemini CLI, Cursor, OpenClaw, OpenCode, Hermes, Pi. Mix and match per agent. Use your existing API keys. 30+ integrations. Stripe, Vercel, Resend, HubSpot, Twilio, Namecheap, Cloudflare, Supabase, Slack, Linear, GitHub, and more. Each one lands as MCP tools that your agents can call inside the budgets and workflows you define.Who it is for
Four shapes of team use Company Agents today: Solo founders who want to turn one person into a full team. One human sets direction. Ten agents handle the work of the hires they cannot afford yet. Agencies and studios running multiple clients in parallel. Each client is a first-class object with its own projects, workflows, and budget. Humans own relationships and creative direction. Agents handle research, drafts, builds, and QA across every project simultaneously. AI-native startups building a real company with one human and twenty AI workers. Agents run engineering, content, ops, and support. The founder runs strategy. Engineering teams adding AI teammates to an existing team. They ship PRs, page humans on review and risky ops, stay audited and budgeted at every turn.What Company Agents is not
It is not a project management tool. It is not a task tracker. It is not a Linear or Asana competitor. It is not a chat interface for a model. It is not a “connect your OpenAI key and get prompts” product. It is the thing you run your whole AI-staffed company inside.Next
- Quickstart gets you from install to your first agent run in five minutes.
- Core concepts walks through the six primitives (teams, clients, integrations, budgets, memory, audit) that shape every other page of these docs.
- Architecture is the mental model for what runs where and how the pieces talk to each other.