Other tools and compatibility

Tokens can serve any tool that lets you set a custom OpenAI-compatible or Anthropic-compatible endpoint. This page lists which coding agents and editors can and can't, and gives a generic recipe for tools without their own guide.

Compatibility table#

Based on each tool's official documentation, checked October 2026. "Partial" means some features use Tokens and others don't.

Why Gemini CLI can't use Tokens#

Gemini CLI's only base URL overrides, GOOGLE_GEMINI_BASE_URL and GOOGLE_VERTEX_BASE_URL, expect the Gemini or Vertex request format. Tokens doesn't serve that format, so no setting makes it work. Qwen Code, originally a Gemini CLI fork, supports OpenAI-compatible endpoints.

Connect any OpenAI-compatible tool#

If your tool isn't listed, look for a setting called "OpenAI compatible", "custom provider", "API base" or "base URL". Then you need four things.

1. Base URL#

Some tools, such as Goose's custom provider file, want the full endpoint https://tokens.bd/v1/chat/completions. If the tool's own example ends in /chat/completions, copy that shape. A 404 on the first request almost always means /v1 is missing or doubled.

2. API key#

Tokens reads the key from Authorization: Bearer <key> or x-api-key: <key>, so OpenAI-style and Anthropic-style tools both work without extra headers. Where the tool reads environment variables, use export TOKENS_API_KEY=tok_live_your_key and keep the key out of config files you might commit. One key per tool makes usage easy to read and revocation painless; see API keys.

3. Model ID#

Use the exact Tokens ID, including the provider prefix: deepseek/deepseek-v4.1-flash, not deepseek-v4.1-flash. Get IDs from the model catalog or from the API, which returns only models your key can use:

curl -s https://tokens.bd/v1/models -H "Authorization: Bearer $TOKENS_API_KEY"

Tools that reference models as <provider>/<model> end up with two slashes, such as tokens/deepseek/deepseek-v4.1-flash. Not every tool documents how it splits that, so check there first if the model isn't found. LiteLLM-based tools such as Aider want an openai/ prefix, which they strip before sending.

4. Context window and output limit#

Many tools can't read a model's limits from the endpoint, so they ask you or assume a default. Use the values from the catalog. A wrong context window breaks automatic compaction: too high and long conversations get rejected upstream, too low and history is trimmed early.

Also check:

• Tool calling. Agents that edit files need a model with function calling.
• Streaming. For a usage chunk at the end of an OpenAI-style stream, the client must send stream_options: {"include_usage": true}.
• Unsupported endpoints. Images, audio, files, batches, assistants, fine-tuning and moderations return 404 unsupported_endpoint. Embeddings work only for embedding models, so keep a tool's embedding features on their current provider.
• Browser-only tools. Tokens sends no CORS headers, so calls straight from a web page fail. Use a server, CLI or editor extension.

Test the connection#

Before you debug a tool's settings, confirm the key and model work. The tester below sends a short request with your key and the model you pick. Or run the curl that follows from your machine.

curl -s https://tokens.bd/v1/chat/completions \
  -H "Authorization: Bearer $TOKENS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "deepseek/deepseek-v4.1-flash", "max_tokens": 16, "messages": [{"role": "user", "content": "Reply with OK"}]}'

If that succeeds and the tool still fails, the problem is in the tool's settings. Connect Your Agent also runs a live test and has snippets for common agents.

Troubleshooting#

Every code is in Troubleshooting; formats are in Chat Completions and Messages. The Tokens CLI can configure OpenCode, Claude Code, Codex CLI and Crush for you.

Sources: Gemini CLI configuration reference, MoonshotAI/kimi-cli archive notice, plus the official docs linked from each tool's guide, checked October 2026.