Governed agentic coding.
Bring your own coding agent — Claude Code, Cursor, Cline, Aider, or any OpenAI/Anthropic-compatible tool — and route it through Anoman's guarded gateway. One install script, no code changes, fully revertible.
Install
One command. Your existing API key.
The script points your coding agent's base URL at Anoman using your existing dashboard API key, backs up any file it edits (<file>.anoman.bak), and can be fully reverted with --revert.
curl -fsSL https://anoman.io/setup-agent.sh | shPrefer to verify before running an already-downloaded copy? Checksums are published at setup-agent.sh.sha256 and setup-agent.ps1.sha256. The script never executes anything it downloads and your API key is written only to your agent's own config — never logged or sent anywhere else.
Why route your coding agent through Anoman
Every call your agent makes gets the same guardrail pipeline
Reversible PII masking of source & secrets
Emails, keys, and other detected PII in prompts and diffs are anonymized before they reach the model, then restored in the response your agent sees. The model never sees the real values.
Prompt-injection defense
Every request — including tool output your agent feeds back into the model — passes through injection detection before it's routed, catching malicious instructions hidden in files or fetched content.
Tool-call policy & org governance
Allow or deny specific tool calls per key or per team, and see every agent's activity in one place — useful the moment more than one person on your team is running an agent against your codebase.
Cost arbitrage on every call
Batch routing and prompt caching fund below-market pricing on the same models your agent already calls — you don't change how your agent works, just what it costs.
Data residency, Indonesia-ready
Anoman's Jakarta region keeps prompts and traces in-region for UU PDP readiness. Need prompts to never leave your own network at all? The Edge Collector adds on-prem residency (Enterprise, early access).
Same agent, same workflow
Nothing about how you use your agent changes — no new CLI, no new SDK, no code to rewrite. It's the same tool, pointed at a different base URL.
Runaway-loop protection
Attach a session id and Anoman enforces a per-session step, token, and cost ceiling — so a misbehaving agent loop trips a hard stop in seconds instead of silently burning your budget on an endless retry.
Supported agents
Full setup guides
Each of these has a step-by-step integration guide with copy-paste configuration.
The installer's interactive picker also configures Codex CLI, OpenCode, and Goose directly — no separate guide yet, but running the script walks you through it.
Also works with agent frameworks
See the full integrations page for every OpenAI- and Anthropic-compatible option, or browse the complete agent guide index.
Honest limitation: the installer only works with agents that let you point them at a custom endpoint. Locked, single-provider assistants — GitHub Copilot, Amazon Q, JetBrains AI Assistant — don't expose a custom base URL, so they can't be routed through Anoman.
How it works
What the script does, step by step
1. Get your API key from the Anoman dashboard (Dashboard → API Keys)
2. (Optional) verify the published SHA-256 checksum
3. Run the one-line installer for your OS
4. Pick your agent from the interactive menu, paste your key
5. Launch your agent — it now routes through Anoman's guarded gateway
↳ Undo any of it any time: sh setup-agent.sh --revert
FAQ
Common questions
Which operating systems does the installer support?
macOS and Linux via curl -fsSL https://anoman.io/setup-agent.sh | sh, and Windows via PowerShell with irm https://anoman.io/setup-agent.ps1 | iex. Both are plain shell/PowerShell scripts — no separate binary to install.
Which coding agents does it work with?
Claude Code, Cursor, Cline, Continue, Roo Code, Windsurf, Aider, and Zed all have a dedicated setup guide. The installer's interactive picker also configures Codex CLI, OpenCode, and Goose directly. Any agent or SDK that accepts a custom OpenAI- or Anthropic-compatible base URL can be pointed at Anoman manually, even without a guide.
What does the script actually change on my machine?
It sets your agent's base-URL environment variable (or edits the agent's own config file, e.g. Continue's config.json) to point at Anoman instead of the provider directly, using the API key you paste in. Any file it edits is backed up first as <file>.anoman.bak, and it never executes anything it downloads.
Is my source code sent to Anoman?
Only what your agent already sends as part of a normal request — the same prompt and file context it would send straight to OpenAI or Anthropic. Anoman's gateway then screens that traffic for prompt injection and can mask PII and detected secrets in transit. The install script itself doesn't read, upload, or scan your codebase — it only edits configuration.
How do I undo it?
Run the same installer with --revert (sh setup-agent.sh --revert, or the .ps1 equivalent on Windows). It restores every file it backed up to its original contents. Nothing is left dangling.
Does it need my password or a new account?
No. You paste in the API key you already have from your Anoman dashboard (or pass --api-key, or set ANOMAN_API_KEY beforehand). There's no separate login step and the script never asks for a password.
Your customers are already asking. Have the answer ready.
Ship AI you can prove is safe — guarded, logged, and Jakarta-resident on every call. Founding 50 seats and 50%-off launch pricing are limited.