Guild.ai vs Orchestrator Zero
oz0 records agent jobs running on your infrastructure. Guild offers hosted agent infrastructure and proxy-based governance. oz0 provides a control plane you host, with outbound-only nodes, reviewed tool definitions and a verifiable audit trail.
What Guild.ai is
Guild.ai is a hosted control plane for AI agents, in open beta since April 2026. Teams build agents with its TypeScript SDK or bring Goose, OpenClaw and LangGraph agents, and Guild runs them on Google Cloud. Every model and tool call goes through a Guild proxy that holds the credentials, applies policies and budgets, and records the session.
docs.guild.ai- Runs
- Managed service
- Checked
- 27 October 2026
Side by side.
Guild.ai's ratings come from its own pages. A column's name opens its definition and how Orchestrator Zero does it. How we rated
Where it runs
Any machine you own
Yes
DocsNo: Runs in Guild's runtime
Agent code runs in Guild's own runtime; the docs describe no way to run agents on machines you choose.
Linux, macOS and Windows
No: Guild's cloud only
Agents run only in Guild's hosted runtime; there is no runtime for your own Linux, macOS or Windows machines.
Self-hosted control plane
Yes
DocsNo: Hosted on Google Cloud
Guild is a hosted service on Google Cloud; no self-hosted, VPC or on-premises option is documented.
No inbound ports
Yes
DocsNo: API forwards to the runtime
Guild's API forwards each request to the runtime that runs the agent; no machine of yours runs agents.
Fleet operations
Yes
DocsNo: Guild runs the compute
The runtime loads, runs and manages every agent; the docs have no machines to join, label, route to or block.
Durability
Durable runs
Conditional: TypeScript state machines
TypeScript agents using the state compiler can persist and resume at await boundaries after a runtime restart. Equivalent recovery is not documented for every supported agent runtime.
Governance
Tool policy outside the model
Conditional: Allow and deny, no approve
YAML credential policies allow or deny each outbound call at Guild's proxy, by operation, domain, method, agent and workspace; new credentials start allow-all, and no rule sends a call to a person.
Approvals in chat
Conditional: Agent asks; no call gate
An agent can ask the user a question with ui_prompt and wait for the answer; no rule makes a tool call wait for a person's approval, and no approvals reach a chat app.
Tool snapshots
Yes
DocsConditional: Versioned, no change check
Integrations are versioned packages that agents name by semver range; no check or hold on a changed MCP tool definition is described.
Tenants built in
Yes
DocsYes: Organizations, workspaces
Organizations hold workspaces. A workspace can be restricted to named members and organization admins, although shared agent, integration and runtime dependencies retain their own access rules.
Cost
Budgets that stop a run
Yes: Monthly USD, blocks at 100%
Admins set monthly spend budgets per workspace, agent or user; Guild warns at 80% and blocks further LLM calls at 100%.
Security
Provider keys off the workers
Yes
DocsYes: Credential proxy holds them
Service and model provider credentials are held server-side and added by Guild's proxy, so they never enter the agent's code, container or runtime.
Sandbox with an egress allowlist
Yes: Proxy only, domains by policy
An agent's container reaches only the Guild API, so outbound calls go through Guild's proxy, where credential policies can limit HTTP calls to hostname patterns.
Audit and compliance
Tamper-evident audit trail
Conditional: No documented check
Admins can read audit records of administrative actions and policy decisions, and entries cannot be edited or deleted through the product. A hash chain, signature or independent verification method is not documented.
Erasure by key
Not documented
Searched the docs for deleting a person's data, per-person keys and erasure: nothing found; the privacy policy says to contact Guild.
Retention with legal hold
Not documented
Searched the docs for retention settings and legal hold: nothing found. Audit log entries cannot be modified or deleted.
Openness
Model provider choice
Yes
DocsConditional: Many vendors, no local
Managed Guild tokens, or your own keys for Anthropic, OpenAI, Google AI and Meta, with Bedrock, OpenRouter, Fireworks and Vertex AI when enabled; no local model is documented.
MCP and A2A
Yes
DocsConditional: MCP, no A2A
Agents use MCP integrations, and the CLI serves Guild operations as MCP tools; A2A is not in the docs.
Where Guild.ai is stronger.
Nothing to run yourself
Sign up in the browser with Google, GitHub or an email link and start from chat; Guild runs the control plane and the agents.
docs.guild.aiAn assistant that sets it up
Smith, a built-in agent, builds and publishes agents, sets up integrations and triggers, and shows each policy change for approval before it writes.
docs.guild.aiCompany-wide AI spend
Spend Explorer imports spend from Anthropic, OpenAI and Cursor admin keys, so usage outside Guild shows next to Guild's own.
docs.guild.aiVersion states for agents
Each agent version moves from Saved to Validated, once it builds and matches its typed schemas, to Published, with semver bumps.
docs.guild.aiIntegrations from OpenAPI, and a public API
Custom integrations import their operations from an OpenAPI spec, and the platform has a public API with an OpenAPI spec and a WebSocket event stream.
docs.guild.aiSOC 2 reports listed
Guild's trust centre now lists both SOC 2 Type 1 and Type 2 reports. Access to the reports themselves requires approval.
trust.guild.ai
Full in Guild.ai, partial in oz0: Sandbox with an egress allowlist.
Differences to examine.
Compare the scope, conditions and editions in the source notes.
- Any machine you own
- Linux, macOS and Windows
- Self-hosted control plane
- No inbound ports
- Fleet operations
- Tool snapshots
- Tamper-evident audit trail
- Model provider choice
- MCP and A2A
Using both
A Guild custom integration can point at an MCP server, and every Orchestrator Zero agent is an MCP tool, so a Guild agent could hand a step to one, if the edge is reachable from Guild's cloud.
Based on documentation; this pairing has not been tested.
Find your first task.
Explore an example or read the public Docs preview. Software access is coming soon.