Use OpenRouter in your coding agents

Two endpoints, one key

The openrouter preset knows two base URLs for the same key: https://openrouter.ai/api/v1 for OpenAI Chat Completions and https://openrouter.ai/api for Anthropic Messages. Each agent gets the first protocol in its own preference order that the preset offers, so the table above is also a map of which URL lands in which config file. Droid, for example, prefers Anthropic Messages, while pi, opencode and Crush take the Chat endpoint.

A preset is never probed. Instead, whenever yoink lists OpenRouter's models (in the interactive add, or with yoink probe --preset openrouter), it also checks the key against https://openrouter.ai/api/v1/key. If OpenRouter answers 401 or 403, yoink stops with "OpenRouter rejected the API key" before writing anything. Of the seven presets, OpenRouter is the only one with that check. The scripted yoink add --external --preset openrouter takes your --models as given and never contacts OpenRouter; it only looks the ids up on models.dev.

Add it

Interactively, run yoink add, choose Provider (API key), then OpenRouter, paste the key (the input is hidden), pick models and tick harnesses. From a script, pass the key on stdin so it stays out of your shell history:

yoink add
 
echo "$OPENROUTER_API_KEY" | yoink add --external --name openrouter --preset openrouter \
  --models z-ai/glm-4.7 --connect pi,opencode,crush --default z-ai/glm-4.7 --token-stdin
 
yoink connect openrouter --to qwen,kilo,continue

--connect and yoink connect --to take the harness ids. If one of the configs is tracked by git, the scripted forms refuse to write the key into it unless you add --allow-tracked.

Pick models

The model picker is built from OpenRouter's live list at https://openrouter.ai/api/v1/models, so it shows whatever OpenRouter serves today. Lists longer than 12 entries get a search box. OpenRouter model ids carry the vendor as a prefix, such as z-ai/glm-4.7, and yoink writes them exactly like that into every agent. For each model, yoink looks up the context window, output limit, reasoning and image support on models.dev.

yoink models openrouter
yoink models openrouter --set z-ai/glm-4.7,z-ai/glm-4.6

Either form re-syncs every agent the provider is connected to.

Use it in Claude Code

Claude Code reaches OpenRouter through the Anthropic Messages endpoint. Because Claude Code holds one backend at a time, connecting OpenRouter makes it the active yoink profile and writes the seven managed keys into ~/.claude/settings.json, with ANTHROPIC_BASE_URL set to https://openrouter.ai/api and your default model in every tier. Switching back to a Claude account, for example with yoink work, strips those keys again.

yoink connect openrouter --to claude-code --default z-ai/glm-4.7
yoink work

The older single-model form still works (the macOS app's single-model add uses it). It saves https://openrouter.ai/api as one Anthropic Messages endpoint with one model, ready for Claude Code:

echo "$KEY" | yoink add --external --name glm --provider OpenRouter \
  --base-url https://openrouter.ai/api --model z-ai/glm-4.7 --token-stdin

Rotate the key

Paste a new key with yoink edit openrouter (the API key field), or pipe it in. Saving the change re-writes the key into every connected agent in one step, and yoink scrubs the old key out of the backups it touched along the way.

echo "$NEW_OPENROUTER_KEY" | yoink edit openrouter --token-stdin

FAQ