Use PleumRouter with Any Coding Agent

It's OpenAI-compatible — point your agent at one base_url and you're done.

PleumRouter is an OpenAI-compatible API. Any agent that supports an OpenAI Compatible provider works as-is: set the base_url to https://apirouter.pleum.ai/v1 and the key to your plm_ key. Below are setup snippets for popular agents.

1. Common — base_url and API key#

First sign up, then issue a key on the dashboard's API Keys page (keys start with plm_). Whatever the agent, all you need is the base_url (https://apirouter.pleum.ai/v1) and the key.

environment

export PLEUM_API_KEY=plm_xxxxxxxxxxxxxxxx
export OPENAI_API_BASE=https://apirouter.pleum.ai/v1
# Qwen Code, MiniMax Code, MiMo Code and OpenClaw read OPENAI_BASE_URL instead
export OPENAI_BASE_URL=$OPENAI_API_BASE
# most agents read OPENAI_API_KEY; Codex, OpenCode, Crush and Pi read PLEUM_API_KEY
export OPENAI_API_KEY=$PLEUM_API_KEY

2. IDE · GUI agents#

These agents let you pick OpenAI Compatible (or Custom OpenAI) as the provider in settings and paste the values below — Cline, Roo Code, Kilo Code, Cursor.

Always use the OpenAI Compatible slot. The dedicated OpenRouter entry is hardcoded to openrouter.ai and will fail. Cursor requires /v1 in the base URL and a non-empty key field. Kilo Code: if the model list does not fill in, enter the model ID manually.

3. Terminal · CLI agents#

Terminal agents mostly take an OpenAI-compatible endpoint via environment variables or a config file. Goose and OpenHands need the openai/ prefix on the model name.

Gemini CLI and Antigravity CLI (agy) use Google native /v1beta. GOOGLE_GEMINI_BASE_URL is the origin without /v1beta. See Gemini CLI, Antigravity CLI, and Gemini API.

4. Desktop apps#

Apps you keep in the Dock do not inherit shell env. pleum wire mints a dedicated plm_ key via POST /v1/keys and merges it into ~/.codex/config.toml and ~/.claude/settings.json. pleum launch applies temporary settings to that run only; Desktop reads user-level files. Codex needs supports_websockets = false and Bearer http_headers. Claude Code Desktop uses the in-app Configure Third-Party Inference form — do not put /v1 on the base URL. See the Codex, Claude Code, and Cursor cookbooks.

5. Model IDs and auto-discovery#

Find model IDs at GET /v1/models or on the models page. Continue fetches this list automatically, but some tools, such as OpenCode, need the models listed in their config file. OpenRouter-format IDs (openai/gpt-5.5) are converted automatically.

Claude Code · LiteLLM

list models

curl https://apirouter.pleum.ai/v1/models \
  -H "Authorization: Bearer $PLEUM_API_KEY"

6. Image · audio · video (direct API)#

Image, audio and video are also OpenAI-compatible HTTP endpoints — image POST /v1/images/generations, TTS POST /v1/audio/speech, STT POST /v1/audio/transcriptions, and video POST /v1/video/generations (async — returns a job_id, then poll GET /v1/jobs/{job_id}). Swap MODEL_ID for a model that supports the modality — see the models page or GET /v1/models.

multimodal endpoints

# Image generation
curl https://apirouter.pleum.ai/v1/images/generations \
  -H "Authorization: Bearer $PLEUM_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "model": "MODEL_ID", "prompt": "a red bicycle", "n": 1, "size": "1024x1024" }'

# Text-to-speech (returns audio bytes; cost in X-Cost-Krw header)
curl https://apirouter.pleum.ai/v1/audio/speech \
  -H "Authorization: Bearer $PLEUM_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "model": "MODEL_ID", "input": "Hello there", "voice": "alloy" }' --output speech.mp3

# Speech-to-text (multipart upload)
curl https://apirouter.pleum.ai/v1/audio/transcriptions \
  -H "Authorization: Bearer $PLEUM_API_KEY" \
  -F model=MODEL_ID -F "file=@audio.mp3;type=audio/mpeg"

# Video generation is async: POST returns a job_id, then poll GET /v1/jobs/{job_id}
curl https://apirouter.pleum.ai/v1/video/generations \
  -H "Authorization: Bearer $PLEUM_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{ "model": "MODEL_ID", "prompt": "a drone shot over a forest" }'

7. More agents#

Most agents not listed here work the same way. Via OpenAI-compatible base_url — OpenHands, Open Interpreter, SWE-agent, Qwen Code, MetaGPT, GPT-Pilot, ChatDev, Tabby (chat), Dyad, Plandex, bolt.diy, Forge, Kimi CLI, gptme, Letta, and more. Via Anthropic-compatible (/v1/messages): claude-code-router also connects with ANTHROPIC_BASE_URL. LiteLLM-based agents (Aider, OpenHands, Open Interpreter, SWE-agent, gptme) connect directly via base_url since PleumRouter is already a gateway — but if you keep a LiteLLM proxy, configure it as below (don't append /chat/completions to api_base).

Stuck somewhere? Drop your key into the playground and test it right away. Top-up bonus terms are in the pricing docs.