Conductor × Portkey is now live

In March 2026, Anthropic started throttling Claude Code sessions during peak hours. Max plan users paying $200 a month were burning through their five-hour windows in under 20 minutes. Anthropic confirmed that they were intentionally adjusting session limits to manage growing demand.

Conductor is one of the few tools in the Claude Code ecosystem that actually changes how you work. The git worktree isolation is well thought through, the diff review workflow is clean, and running multiple agents in parallel is meaningfully different from running them sequentially. Charlie and Jackson built it because they use it all day, and you can tell.

While Conductor handles everything above the API layer: the orchestration, the worktree isolation, the diff review, the parallel sessions ; Portkey handles everything below it: costs, logs, fallbacks, and budget limits on every request flowing through those sessions.

The two are meant to work together.

Setup

Since Conductor uses Claude Code under the hood, it accepts the same ANTHROPIC_* environment variables. Set these in Settings → Providers under Claude Code:

ANTHROPIC_BASE_URL=https://api.portkey.ai
ANTHROPIC_AUTH_TOKEN=YOUR_PORTKEY_API_KEY
ANTHROPIC_CUSTOM_HEADERS=x-portkey-api-key: YOUR_PORTKEY_API_KEY
ANTHROPIC_CUSTOM_HEADERS=x-portkey-provider: @anthropic-prod
ANTHROPIC_DEFAULT_SONNET_MODEL=claude-sonnet-4-20250514
ANTHROPIC_DEFAULT_OPUS_MODEL=claude-opus-4-20250514
ANTHROPIC_DEFAULT_HAIKU_MODEL=claude-haiku-4-20250514
ANTHROPIC_API_KEY=

Two things worth noting before you start. Model names are required – without the correct model settings, requests will fail. And ANTHROPIC_API_KEY needs to be set to an empty string explicitly, otherwise Conductor will try to use it and route around Portkey entirely.

What changes?

Observability. Every request in your coding session is logged with full context: request and response, token usage, cost, and latency. You can group requests by session or project using a trace ID header, so logs across parallel workspaces don't blur together. All of it is visible in the Portkey dashboard.

Budget controls. Set hard limits on maximum spend per day, week, or month, on tokens consumed, and on requests per minute or hour. Conductor sessions can run expensive agentic loops. This is how you put a ceiling on them before they run, not after.

Cross-provider fallbacks. Configure automatic failover so a single provider outage doesn't kill your session. The fallback config points to your backup providers in order, and Portkey routes to the next one automatically if the first fails.

3000+ models. Switch providers by changing the provider slug in your config. The same ANTHROPIC_* variables work regardless of what's on the other end. Portkey supports Anthropic, Bedrock, Vertex AI, and many others through the same interface.

Caching. Reduce costs and latency for repeated queries, which are common in iterative coding sessions where the same context goes out across multiple requests

You can find the full setup guide in the docs

If you run into anything, open an issue on the gateway repo or drop a note in the Portkey Discord. We read them :)