What does osr run change in the tool?

Nothing in the tool. It sets the environment variables the tool already reads for its endpoint and key - ANTHROPIC_BASE_URL for Claude Code, OPENAI_BASE_URL for Codex and the rest - and starts the tool with your own arguments. `osr run --print-env` shows the variables for a shell profile instead.

Which model does a tool get?

What it asks for. auto lets the router choose per turn under your workspace policy; a model id pins it; a workspace alias can map the tool's own default name to either, so a tool that insists on gpt-4o still routes.

What applies to the tool's traffic?

Everything that applies to any request: the workspace policy and tenant constraints, the key's budget, rate and allowed targets, PII and secret redaction, the response cache when it is on, and a trace per turn with the alternatives and the policy decisions.

My tool is not listed.

Any tool that speaks the OpenAI chat / Responses API or the Anthropic Messages API works by changing its base URL and key - the migration page shows the before and after for the common SDKs. Ask on the support page and it joins the list.

Integrations

Every coding agent and AI tool the team runs, behind one router.

Claude Code, Codex, Gemini CLI, Aider and OpenCode start with one command; Cursor, Continue, Zed, LibreChat and Open WebUI take a base URL. From then on every turn is routed under your policy, budgeted per key, redacted and traced - and the tool does not know.

Create a free account Application code instead

osr runbash

pip install opensmartroute
osr login                 # browser sign-in once, or --token osr_live_...
osr run claude-code       # Claude Code, routed
osr run codex -- --full-auto "fix the failing tests"
osr run --list            # every tool and how it is wired

Terminal agents

Launched with osr run: the credential becomes the tool's endpoint and key; the rest of the command line is the tool's.

Claude Code
Claude Code reads ANTHROPIC_BASE_URL and ANTHROPIC_AUTH_TOKEN; the router serves /v1/messages.
osr run claude-code
Codex CLI
Codex speaks the Responses API; the router serves /v1/responses on the same key.
osr run codex
Gemini CLI
The router speaks the Gemini API at /v1beta, so the CLI keeps its own dialect and key variable.
osr run gemini-cli
Aider
Aider's OpenAI-compatible mode: OPENAI_API_BASE and a model prefixed with openai/.
osr run aider
OpenCode
Add the router as an OpenAI-compatible provider in opencode.json, or launch with these variables.
osr run opencode

Editors

A base URL and a key in the editor's model settings; the model name auto lets the router choose.

Cursor
Settings > Models > OpenAI API key: paste the key, override the base URL with the router's /v1, add the model name auto.
OPENAI_BASE_URL
Continue
In config.yaml add a model with provider openai, apiBase the router's /v1 and model auto.
OPENAI_BASE_URL
Zed
In settings.json set language_models.openai.api_url to the router's /v1 and list the model auto.
OPENAI_BASE_URL

Chat front-ends

An OpenAI-compatible connection for a whole team, with the router's models in the picker.

LibreChat
A custom endpoint in librechat.yaml with baseURL the router's /v1 and the key.
OPENAI_BASE_URL
Open WebUI
Settings > Connections > OpenAI API: the router's /v1 and the key; auto appears in the model list.
OPENAI_API_BASE_URL

Questions

How the tools and the router fit

What does osr run change in the tool?: Nothing in the tool. It sets the environment variables the tool already reads for its endpoint and key - ANTHROPIC_BASE_URL for Claude Code, OPENAI_BASE_URL for Codex and the rest - and starts the tool with your own arguments. `osr run <tool> --print-env` shows the variables for a shell profile instead.
Which model does a tool get?: What it asks for. auto lets the router choose per turn under your workspace policy; a model id pins it; a workspace alias can map the tool's own default name to either, so a tool that insists on gpt-4o still routes.
What applies to the tool's traffic?: Everything that applies to any request: the workspace policy and tenant constraints, the key's budget, rate and allowed targets, PII and secret redaction, the response cache when it is on, and a trace per turn with the alternatives and the policy decisions.
My tool is not listed.: Any tool that speaks the OpenAI chat / Responses API or the Anthropic Messages API works by changing its base URL and key - the migration page shows the before and after for the common SDKs. Ask on the support page and it joins the list.

One router behind every tool.

Create a free account, run osr login once, and every tool on the list follows your policy from its next turn.

Create a free account The MCP server for IDE agents

Free plan, no card. Fifteen thousand decisions a month with the full trace.

pip install opensmartroute osr login # browser sign-in once, or --token osr_live_... osr run claude-code # Claude Code, routed osr run codex -- --full-auto "fix the failing tests" osr run --list # every tool and how it is wired