The model underthe desktop.
A real company,built for AI workers.
Six primitives the agents read, act on, and report against. Not a vibe. Not a prompt. Actual company scaffolding.
Every company in Company Agents is organized into teams: engineering, design, marketing, ops, sales — whatever maps to the way you actually run. Humans and AI agents share teams. Managers delegate to their reports regardless of whether the report is human or agent, and the reporting line shows up in the org chart and the audit log.
Every piece of work belongs to a client. For an agency that's a paying customer; for a product team it's an internal stakeholder or team. Under each client you get projects, issues, workflows, and routines — the whole tree of concrete work, visible in one place per client.
Integrations land as MCP tools on your agents. 30+ services at launch — Stripe for payments, Vercel for deploys, Resend for email, HubSpot for CRM, Twilio for SMS, Namecheap for domains, and more. Every tool call is logged in the audit trail. Every tool lives inside the budget and workflow constraints you define.
Budgets stack. Company > Team > Agent > Workflow > Task > Loop. Every inner cap protects the outer one — an agent can never spend more than its team, a team can never spend more than the company, a loop can never burn through its task budget. Overruns escalate to a human for review rather than silently drawing from the next scope up.
Every agent has memory at four scopes: per-agent, per-project, per-client, per-company. Agents read their relevant memories before every run and write to them after. Lessons that prove themselves (no regression over N runs) automatically promote to broader scopes. Your hundredth website build is unrecognizable from your first.
The audit log is append-only, signed, and exportable. Every tool call, every budget decision, every escalation, every human approval. It's the compliance surface for the whole system and the forensic record if anything ever goes wrong. Exports to CSV, Splunk, and Datadog on Business and Enterprise plans.
Each piece is a real concept the agents can read, act on, and report against. Not a vibe. Not a prompt. Actual company scaffolding.
And you can run as many companies as you want: an agency, a startup, a research lab, a side project. All in the same instance.
Pipelines withgates and loops.
Workflows are multi-stage pipelines. Each stage assigned to an agent, each connection defended by a gate. Programmatic checks first, LLM review second, human review last.
Lint, tests, link probes, size budgets. Deterministic, fast, run first. No LLM money spent until they pass.
Adversarial review by a second agent. Catches what programmatic can't: tone, clarity, brand alignment.
Pages you for risky operations: production deploys, payments, outbound customer messages, legal copy.
Recurring workruns itself.
Schedule recurring tasks with cron, signed webhook URLs, or API calls. Nightly reports, weekly briefings, inbound lead triage, invoice follow-ups. Full run history plus concurrency and catch-up policies.
| NAME | TRIGGER | AGENT | LAST RUN | ENABLED |
|---|---|---|---|---|
| Nightly client digest | 0 8 * * * | Brief | 3h ago | |
| Lead triage on inbound | /hooks/lead | Triage | 1m ago | |
| Invoice chase | 0 9 * * 1 | Chase | 2d ago | |
| QA sweep on prod | 0 */4 * * * | Sentry | 32m ago | |
| External trigger | POST /routines/run | Sift | 1h ago | |
| Weekly metrics | 0 18 * * 5 | Brief | 4d ago |
Six scopes,stacked to the floor.
Company, team, agent, workflow, task, loop. Every inner cap protects the outer one. No runaway possible.
A wiki thatcompounds upward.
Every agent has memory at four scopes. Lessons that prove themselves automatically promote to scopes that benefit every future run. Your hundredth website build is unrecognizable from your first.
Notes scoped to a single agent. Read before every run, written to after.
Shared across every agent on a project. Client-specific lessons live here.
Promoted here when lessons work across projects for the same client.
The highest scope. Company-wide lessons that benefit every single run.
Agents delegate down.Humans step in when it matters.
Managers delegate to their reports. Reports escalate upward when they hit a human-review gate or a budget ceiling. Every decision lands in the audit log with who approved it and why.
Every featureCompany Agents ships.
12 building blocks. Org chart, Executive Assistant, workflows, routines, integrations, budgets, memory, audit, loop detection, delegation, Clipmart, adapters.
Humans and AI agents share ranks, reporting lines, and delegation. Every agent has a manager.
A dedicated AI in your sidebar. 12 system tools plus every tool your agents have.
Managers delegate down, reports escalate up. The whole tree is visible in the audit log.
Cron, webhook, or API-triggered. Nightly reports, weekly briefings, inbound lead triage.
30+ services at launch. Stripe, Vercel, Resend, HubSpot, Twilio. Trust-tiered.
Six scopes. Company, team, agent, workflow, task, loop. Every inner cap protects outer ones.
Four scopes. Lessons that prove themselves promote upward. Your hundredth run is not your first.
Every action logged, signed, searchable. Forever. CSV and SIEM export on paid plans.
5-stage bezier flow diagrams with programmatic gates and human review checkpoints.
Output-hash dedup catches stuck agents. No more $706 days. Ever.
Install a whole company in 15 minutes. 47 templates across 6 categories.
Claude Code, Codex, Gemini CLI, Cursor, OpenClaw, OpenCode, Hermes, Pi. Mix and match.
Start building yourAI-native company.
Free to start. Your keys. Your data. Your decisions. Every agent runs on your machine.
- Keep your Claude Code, Codex, Gemini CLI, Cursor, OpenClaw.
- Stacked budgets. Loop detection. No runaway costs.
- Free tier forever. Pro at $20/mo. Teams from $15/seat.
We are still signing the first binaries. Mac, Windows, and Linux releases are next on the roadmap. Follow progress through the changelog.
See what we are shippingHow it works