Comparison
Relay vs Factory
Factory is a platform for an organisation — droids reachable from Slack, Linear and the browser, with airgapped deployment for enterprises. Relay is one binary a developer installs. The gap is org surface area, not agent quality.
Facts checked against official docs on 2026-09-08
What Factory is
Factory is an agent-native software development platform whose agents ("Droids") run across a desktop app, the Droid CLI, web and mobile at app.factory.ai, Slack and issue trackers, plus Droid Computers — persistent compute environments that retain state across sessions, in Bring-Your-Own-Machine and managed variants. First-party SDLC integrations are GitHub, GitLab, Linear and Slack, with a Connectors catalog offering managed authentication for Notion, Sentry and Google Workspace, and MCP for anything else. Its privacy docs state Factory "does not upload or index your codebase into a remote datastore".
Pricing: Pro $20/mo · Plus $100/mo (about 5× Pro usage, plus Droid Computers) · Max $200/mo (about 10× Pro usage) · Teams $60/mo per team plus $40/mo per seat, up to 10 seats, with 10 hrs/mo of shared Droid Computers · Business and Enterprise custom. The pricing page lists no free plan, though factory.ai/oss offers free access to open-source contributors by application. Bring-your-own-key usage carries a free allowance and is metered after it. As of 2026-09-08.
Side by side
| Factory | Relay | |
|---|---|---|
| Bring your own key | Yes — three provider types (Anthropic Messages, OpenAI Responses, and a generic chat-completions shape) with documented configs for OpenRouter, Fireworks, Groq and others. Keys stay in a local settings file. BYOK works only in the CLI and desktop app, and Factory meters BYOK usage after a free allowance. | Yes, unmetered by Relay. 31 provider namespaces plus OAuth to an existing Claude, Copilot or ChatGPT subscription. |
| Account to run the CLI | Yes for the publicly distributed CLI — the quickstart signs you in through a browser. The airgapped enterprise deployment is the documented exception. | No. Keyless install, and the CLI never checks a licence. |
| Where your code goes | Not uploaded or indexed — no cold copy in Factory cloud. Session content does reach Factory in the cloud-managed pattern; the airgapped deployment makes Factory cloud unreachable at runtime. | Nowhere. The harness runs locally and calls your provider. |
| Autonomy control | Levels Off / Low / Medium / High, set per session or with `droid --auto high`; each command and MCP tool carries a risk level and runs automatically only at or below the chosen level. | Budgets and gates rather than levels: turn, token and wall-clock limits, plus completion gates that are your shell commands, retried a set number of times. |
| Enterprise deployment | Cloud-managed, hybrid, and a documented fully airgapped pattern with no runtime connectivity to Factory. SSO, SAML/SCIM, audit logging and Zero Data Retention on Business and above. Org policy can forbid user-supplied BYOK. | None of this exists. Relay has no SSO, no SCIM, and no audit of which engineer ran which agent. |
| Where work can start | A Linear ticket, a Slack mention, a GitHub event, the web app, or the CLI. | A terminal on a developer’s machine. |
Where Factory is the better tool
- Work that starts outside a terminal — a PM files a ticket or someone tags the bot in Slack and a droid picks it up asynchronously.
- Execution that survives the laptop. Droid Computers persist across sessions; Relay’s persistence stops at the machine.
- The identity and compliance surface Relay’s own pricing page concedes it does not have: SSO, SAML/SCIM, audit logging, Zero Data Retention, EU data residency.
- The ability to forbid bring-your-own-key. An organisation that must route all traffic through an approved gateway can enforce that; Relay has no mechanism to stop a developer using any key they like.
- A documented fully airgapped deployment, which Relay does not have as a product — only as a consequence of being local.
- Managed connectors with guided authentication, against Relay’s configure-it-yourself MCP.
The honest take
The framing to avoid here is "your code stays local with Relay" — Factory’s own docs say it does not upload or index your codebase either, so that is not a differentiator and claiming it would be the kind of thing a reader checks and holds against you. What is different is the shape of the thing. Factory is buying you an organisation’s worth of surface: ticket-to-PR automation, managed connectors, an admin who can set policy, and an airgapped option for the parts of the world that need one. Relay is buying you one binary with no control plane, which is exactly right for a developer or a small team and exactly wrong for a security review that expects SSO and an audit trail. Price comparison is not like-for-like either: Factory’s $20 floor is a self-serve individual, and their OSS grant makes the real floor zero for some people. Compare what you are actually buying, not the smallest number on each page.
Sources
- https://factory.ai/pricing
- https://docs.factory.ai/cli/getting-started/quickstart
- https://docs.factory.ai/model-independence/byok
- https://docs.factory.ai/enterprise/network-and-deployment
- https://docs.factory.ai/enterprise/privacy-and-data-flows
- https://docs.factory.ai/autonomy-and-safety/auto-run
Run the harness yourself — one command, your own model key, no account: curl -fsSL https://relayevals.com/harness/install.sh | bash. Quickstart · What is a coding agent harness? · Pricing