OpenRouter

InternAgent can use OpenRouter as a first-class model provider for the main multi-agent discovery pipeline. OpenRouter uses an OpenAI-compatible API, so the same agent code can route requests to models from OpenAI, Anthropic, Google, and other providers through one endpoint.

1. Get an API key

Create an OpenRouter account, add credits, and create an API key from:

https://openrouter.ai/settings/keys

2. Configure InternAgent

Copy the environment template and set your OpenRouter key:

cp .env.example .env

In .env:

OPENROUTER_API_KEY=sk-or-...
OPENROUTER_BASE_URL=https://openrouter.ai/api/v1
OPENROUTER_APP_NAME=InternAgent
OPENROUTER_SITE_URL=https://discovery.intern-ai.org.cn

The dedicated provider config is available at config/openrouter_config.yaml. It sets:

models:
  default_provider: "openrouter"
  openrouter:
    model_name: "moonshotai/kimi-k2.6:free"

You can replace model_name with any OpenRouter model ID that supports chat completions.

3. Run a discovery smoke test

AutoDebug is the smallest built-in task and does not require external datasets:

python launch_discovery.py \
    --config ./config/openrouter_config.yaml \
    --task AutoDebug \
    --exp_backend claudecode

The --exp_backend claudecode experiment backend still requires its own Anthropic/Claude Code setup. OpenRouter is used by InternAgent's main agent model provider layer.

4. Use OpenRouter from the default config

Instead of using config/openrouter_config.yaml, you can edit config/default_config.yaml:

models:
  default_provider: "openrouter"

The openrouter provider block is already included in the default config.

5. Deep Research QA mode

launch_qa.py currently uses the Deep Research workflow's OpenAI-compatible client directly. To route QA calls through OpenRouter, set the OpenAI-compatible environment variables to OpenRouter:

OPENAI_API_KEY=$OPENROUTER_API_KEY
OPENAI_API_BASE_URL=https://openrouter.ai/api/v1
OPENAI_BASE_URL=https://openrouter.ai/api/v1

Then run:

python launch_qa.py --question "What are recent advances in memory-augmented LLMs?"

Troubleshooting

• Authentication failed: check that OPENROUTER_API_KEY is set in .env.
• Unsupported model: confirm the model ID exists in the OpenRouter model list.
• Unsupported model provider: openrouter: make sure your checkout includes internagent/mas/models/openrouter_model.py and the updated model_factory.py.