Does Claude Code need a plugin or a fork?

No. Claude Code reads ANTHROPIC_BASE_URL and ANTHROPIC_AUTH_TOKEN; the router serves /v1/messages. The router is a standard endpoint; the tool is unchanged.

Which model does the tool get?

Whatever it asks for. The name auto lets the router choose per turn under your policy; a model id pins it and still routes among its providers; a workspace alias maps the tool's own default name to either.

What does the router add that the tool does not have?

A per-key budget and rate limit for every developer's Claude Code, enforced before the provider is called. Secrets and personal data pasted into a prompt redacted before they leave the network. The whole session on the activity page with cost per turn; the model that answered on every reply.

Can I use my own provider keys?

Yes - connect them under Model providers and the router executes on your accounts; model usage is billed by the provider, the platform meters requests.

Claude Code through the router

Run Claude Code through the router. Every turn priced, governed and on the record.

Point Claude Code at OpenSmartRoute with one command: the Messages API it speaks is served natively, every turn is routed under your workspace policy, budgets and key limits apply, and each decision carries its trace. Keep Claude, or let auto pick the model the turn deserves.

A per-key budget and rate limit for every developer's Claude Code, enforced before the provider is called
Secrets and personal data pasted into a prompt redacted before they leave the network
The whole session on the activity page with cost per turn; the model that answered on every reply

Create a free account Claude Code docs

One commandbash

pip install opensmartroute
osr login                      # once: a browser sign-in, or --token osr_live_...
osr run claude-code          # Claude Code starts with the router as its endpoint

Or by hand - claudebash

export ANTHROPIC_BASE_URL=https://api.opensmartroute.ai
export ANTHROPIC_AUTH_TOKEN=$OSR_API_KEY
export ANTHROPIC_MODEL=auto
claude

Claude Code speaks the Anthropic Messages API; the router serves it at /v1/messages with tools, images, streaming and extended thinking translated both ways.

Verify

The first turn is on the activity page.

Send one message from the tool and read the row: the model that answered, the cost, the confidence and a link to the full trace with the alternatives and the policy decisions.

verify.shbash

# the tool's next turn is on the activity page with its cost, model and trace
curl -s https://api.opensmartroute.ai/api/v1/activity?limit=3 -H "Authorization: Bearer $OSR_API_KEY"

Questions

Claude Code and the router

Does Claude Code need a plugin or a fork?: No. Claude Code reads ANTHROPIC_BASE_URL and ANTHROPIC_AUTH_TOKEN; the router serves /v1/messages. The router is a standard endpoint; the tool is unchanged.
Which model does the tool get?: Whatever it asks for. The name auto lets the router choose per turn under your policy; a model id pins it and still routes among its providers; a workspace alias maps the tool's own default name to either.
What does the router add that the tool does not have?: A per-key budget and rate limit for every developer's Claude Code, enforced before the provider is called. Secrets and personal data pasted into a prompt redacted before they leave the network. The whole session on the activity page with cost per turn; the model that answered on every reply.
Can I use my own provider keys?: Yes - connect them under Model providers and the router executes on your accounts; model usage is billed by the provider, the platform meters requests.

Other tools

The same router behind every tool the team runs.

Put Claude Code behind the router today.

A free account, one key, one command. Every turn the tool makes is priced, governed and explained from then on.

Create a free account Migrating application code instead?

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