Table of Contents

OpenRouter

OpenRouter is a unified API that routes requests to AI models from multiple vendors — OpenAI, Anthropic, Google, Meta, Mistral, and more — through a single endpoint. JD.AI integrates with OpenRouter as a first-class provider, automatically discovering the full model catalog and importing pricing and capability metadata.

Quick start

  1. Get an API key from openrouter.ai/keys.

  2. Configure JD.AI:

    # Environment variable
export OPENROUTER_API_KEY=sk-or-...

# Or interactive setup
/provider add openrouter

  3. Launch JD.AI — models are discovered automatically:

    Detecting providers...
  ✓ OpenRouter: 247 model(s)

How it works

OpenRouter acts as a proxy that routes your requests to the underlying vendor APIs (OpenAI, Anthropic, Google, etc.). JD.AI communicates with OpenRouter using the OpenAI-compatible chat completions format through https://openrouter.ai/api/v1.

On startup, JD.AI fetches the full model catalog from the OpenRouter API, extracts metadata for each model, and makes them available through /models and /model. When you send a message, JD.AI forwards the request to OpenRouter, which routes it to the appropriate vendor.

Model IDs

OpenRouter uses a vendor/model naming convention:

Example ID Vendor Model
anthropic/claude-sonnet-4 Anthropic Claude Sonnet 4
openai/gpt-4.1 OpenAI GPT-4.1
google/gemini-2.5-pro Google Gemini 2.5 Pro
meta-llama/llama-4-maverick Meta Llama 4 Maverick
mistralai/mistral-large Mistral Mistral Large

Use the full vendor/model ID when selecting a model:

/model anthropic/claude-sonnet-4

Or pass it on the command line:

jdai --model anthropic/claude-sonnet-4

Model discovery

JD.AI fetches the full model catalog by calling GET /api/v1/models with your API key on startup. Each model entry includes context length, max output tokens, pricing, supported parameters, and input/output modalities.

The discovery process:

  1. Sends a GET request with Bearer token authentication
  2. Deserializes the catalog response
  3. Filters to text-capable models (excludes image-only output models)
  4. Extracts metadata and capabilities for each model

When discovery fails — due to network issues, an invalid API key, or a timeout — JD.AI falls back to five built-in models (see Fallback models).

Tip

Use /models to browse all discovered models, filter by name, and see capability badges.

Capabilities and metadata

JD.AI extracts the following metadata for each discovered model:

Field Source Default
Context window context_length 128,000 tokens
Max output tokens top_provider.max_completion_tokens 16,384 tokens
Input cost per token pricing.prompt 0
Output cost per token pricing.completion 0
Tool calling supported_parameters contains tools or tool_choice —
Vision architecture.input_modalities contains image —

Capability badges appear in the /models list:

  • Chat — all models include this
  • Tool Calling — model supports function/tool calling via structured schemas
  • Vision — model accepts image inputs

Pricing

OpenRouter charges per token with prices set individually per model. JD.AI imports these prices so /cost shows accurate estimates for your session.

Example pricing (varies — check openrouter.ai/models for current rates):

Model Input (per 1M tokens) Output (per 1M tokens)
anthropic/claude-sonnet-4 $3.00 $15.00
openai/gpt-4.1 $2.00 $8.00
google/gemini-2.5-pro $1.25 $10.00
meta-llama/llama-4-maverick $0.20 $0.60
mistralai/mistral-large $2.00 $6.00
Note

Prices are fetched from the OpenRouter API at startup. The values above are examples and may change. Use /cost during a session for real-time cost tracking.

Choosing a model

Goal Recommended models Why
Best reasoning anthropic/claude-sonnet-4, openai/gpt-4.1 Strong at complex code generation and analysis
Fastest responses google/gemini-2.5-flash, meta-llama/llama-4-maverick Optimized for low latency
Lowest cost Small open-source models (Llama, Mistral Small, Qwen) Fraction of the cost of large models
Largest context google/gemini-2.5-pro, openai/gpt-4.1 1M+ token context windows
Tool calling anthropic/claude-sonnet-4, openai/gpt-4.1 Reliable structured output and function calling

Browse the full catalog with /models and filter by name to find models matching your needs.

Fallback models

When live discovery is unavailable (network issues, invalid key, timeout), JD.AI uses five built-in models:

Model ID Display name Context Max output
anthropic/claude-sonnet-4 Claude Sonnet 4 200K 16K
openai/gpt-4.1 GPT-4.1 1M 32K
google/gemini-2.5-pro Gemini 2.5 Pro 1M 64K
meta-llama/llama-4-maverick Llama 4 Maverick 1M 1M
mistralai/mistral-large Mistral Large 128K 128K
Note

Fallback models do not include pricing metadata. /cost estimates will not be available until discovery succeeds.

Configuration

Setting Value
Environment variable OPENROUTER_API_KEY
Credential store path ~/.jdai/credentials.enc
API endpoint https://openrouter.ai/api/v1
Discovery timeout 10 seconds
Request timeout 10 minutes

Credential resolution chain (first match wins):

  1. CLI flags — --api-key
  2. Environment variable — OPENROUTER_API_KEY
  3. Encrypted credential store — ~/.jdai/credentials.enc

Troubleshooting

OpenRouter shows unavailable

Check that your API key is set and valid:

echo $OPENROUTER_API_KEY

Re-run the setup wizard to re-enter your key:

/provider add openrouter

Only 5 models listed

Discovery failed and JD.AI is using the fallback list. Common causes:

  • Invalid or expired API key — regenerate at openrouter.ai/keys
  • Network issues — check internet connectivity
  • Timeout — the discovery request has a 10-second timeout; try again

Once the issue is resolved, restart JD.AI or run /provider test openrouter to retry.

Model not found

Use the full vendor/model ID — for example, anthropic/claude-sonnet-4 not just claude-sonnet-4. Browse available models with /models.

Unexpected pricing or capabilities

Pricing and capabilities are fetched from the OpenRouter API at startup. If values seem stale, restart JD.AI to re-fetch the catalog.

See also

Edit this page