Agent Composer
Define models, tools, memory, knowledge, policies, budgets and triggers in one versioned agent contract.
Enterprise self-hosted AI operating system
Agents need an operating system.
Build, connect, govern and run AI agents across your infrastructure — without giving an LLM the keys to your business. AutoAgent OS puts identity, policy, security, audit and cost control around every autonomous action.
Self-hosted · Model-agnostic · Linux-native · Air-gapped ready · No credit card for Community
AUTOAGENT CONTROL PLANEEU / healthy
Policy before prompt.
$ agentos deploy research-team
✓ identity verified
✓ policy compiled
✓ tools scoped
✓ secrets brokered
→ runtime/worker-eu-03
→ model/router:auto
→ autonomy: level-3
READY — approval required for external actions
✓ identity verified
✓ policy compiled
✓ tools scoped
✓ secrets brokered
→ runtime/worker-eu-03
→ model/router:auto
→ autonomy: level-3
READY — approval required for external actions
ZERO TRUSTFAST PATHAUDITQUARANTINEMODEL ROUTER
01 / One control surface
From first agent to ten thousand — one governed runtime.
Most agent stacks begin with orchestration and add governance later. AutoAgent OS starts from the opposite direction: every agent, model, tool, secret and action lives inside an explicit control boundary.
Durable execution
Sync, async, streaming, retries, cancellation, checkpoints, replay and long-running workflows that survive worker failure.
Enforcement outside the LLM
Policy, authorization, secrets, network access and sandbox controls are evaluated independently from model reasoning.
Model routing
Route by quality, privacy, latency, region, cost and availability with cascades and approved failover.
AI Operations Center
Trace users → agents → models → tools → data → actions with cost, performance and security signals in one place.
Your infrastructure
Run on Linux, Docker, Kubernetes, bare metal, private cloud, multi-cluster or fully air-gapped environments.
02 / Security architecture
Autonomy without uncontrolled authority.
Live policy decision
An agent can ask. The policy engine decides.
read:customer / EUALLOW
export:restricted_dataDENY
payment:€12,4002 APPROVERS
tool:unknown_connectorQUARANTINE
Security plane
Zero trust as product UX.
RBAC + ABAC, just-in-time permissions, DLP, secret broker, sandboxing, egress policies, risk scoring, trust scoring, four-eyes approval and an emergency kill switch are designed as operating controls — not prompt instructions.
03 / Where it fits
One platform. Different autonomy boundaries.
AI platform teams
Give developers a standard runtime, model gateway and policy layer instead of rebuilding agent infrastructure for every product.
Regulated enterprise
Keep sensitive workloads in-region or fully internal while preserving audit trails, approval gates and data classification policies.
Operations automation
Run durable workflows that can call tools, wait for humans, recover from failures and remain observable from start to finish.
Private AI
Connect internal models, private knowledge and restricted tools without requiring a single cloud-model vendor or public SaaS control plane.
04 / Pricing
Start locally. Pay when the organization depends on it.
Infrastructure and model-provider costs are separate. Annual plans include roughly two months free.
Community
€0
forever · evaluation & personal use- 1 workspace
- Up to 5 active agents
- Core runtime + workflows
- Model & tool connectors
- Community updates
Team
€199
/month · €1,990 annual- Up to 10 users
- 50 active agents
- Policy engine + audit
- Cost controls
- Email support
- Production license
Business
€599
/month · €5,990 annual- Up to 50 users
- 500 active agents
- SSO + advanced RBAC/ABAC
- Private registry
- Security operations
- Priority support
Enterprise
Custom
contract · SLA · deployment services- Unlimited / negotiated scale
- Multi-cluster & multi-region
- Air-gapped deployment
- Data residency controls
- Enterprise support + SLA
- Architecture onboarding
Give every agent a boundary.
Create workspace
No credit card required.