Connect Zed to Tokens

Zed is a fast code editor with a built-in agent, the Zed Agent. It connects to Tokens through an OpenAI-compatible provider declared under language_models.openai_compatible, which sends Chat Completions requests to https://tokens.bd/v1. Zed can also use an Anthropic-compatible provider; that variant is covered near the end.

This guide was checked against Zed v1.22.0 (October 2026).

Add the provider from the Agent Panel#

The quickest route is the UI:

1. Open Agent Settings. From the command palette, run agent: open settings (action agent::OpenSettings).
2. Under LLM Providers, click Add Provider.
3. Enter the provider name Tokens, the API URL https://tokens.bd/v1, the model ID deepseek/deepseek-v4.1-flash, and the context window from the model catalog.
4. Enter your key, tok_live_your_key, in the provider's key field.

If you don't have a key yet, create one in the dashboard. See API keys.

Declare Tokens in settings.json#

For a reproducible setup, declare the provider in your Zed settings.json (on macOS and Linux, ~/.config/zed/settings.json):

{
  "language_models": {
    "openai_compatible": {
      "Tokens": {
        "api_url": "https://tokens.bd/v1",
        "available_models": [
          {
            "name": "deepseek/deepseek-v4.1-flash",
            "display_name": "DeepSeek V4.1 Flash",
            "max_tokens": 128000,
            "max_output_tokens": 8192,
            "capabilities": {
              "tools": true,
              "images": false,
              "parallel_tool_calls": false,
              "prompt_cache_key": false,
              "chat_completions": true,
              "interleaved_reasoning": false,
              "max_tokens_parameter": true
            }
          }
        ]
      }
    }
  }
}

max_tokens here is the context window and max_output_tokens the output limit. The values shown are examples; use the numbers listed for the model in the catalog.

Provide the key#

Zed builds the environment variable name from the provider name: <PROVIDER_NAME>_API_KEY. For a provider called Tokens, that's TOKENS_API_KEY.

export TOKENS_API_KEY=tok_live_your_key

Zed must be started from an environment that has this variable. If you launch Zed from a dock or start menu, it may not see variables exported in your shell profile. Entering the key in the Agent Panel avoids that problem.

Capability flags#

Zed's defaults are tools: true, images: false, parallel_tool_calls: false, prompt_cache_key: false, chat_completions: true, interleaved_reasoning: false and max_tokens_parameter: false. Three are worth thinking about:

• max_tokens_parameter: true makes Zed send max_tokens instead of max_completion_tokens. max_tokens is the classic Chat Completions field and the safer choice for a gateway serving many providers, which is why the example turns it on.
• interleaved_reasoning: true sends earlier thinking back in reasoning_content, which DeepSeek-style reasoning models use. Leave it off unless the model you chose is a reasoning model that needs it.
• chat_completions: false switches Zed to the Responses API. Keep it true. Tokens serves /v1/responses, but support depends on the upstream for each model, while Chat Completions works for all of them.

Set images: true only for models the catalog lists with image input.

Switch models#

Add more objects to available_models, one per Tokens model, then pick between them in the Agent Panel model picker. Use exact IDs from the catalog or from:

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

Choosing a model explains the trade-offs for agent work.

Anthropic-compatible variant#

Zed also supports language_models.anthropic_compatible. Set api_url to https://tokens.bd (no /v1; the Anthropic protocol adds the path itself). The capability flags for that provider are tools, images and prompt_caching. Zed sends the key in an X-Api-Key header, which Tokens accepts. The OpenAI-compatible setup above is the simpler choice for most models; the Messages API is described in Messages.

Verify it works#

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 works, select "DeepSeek V4.1 Flash" in the Agent Panel model picker and send a prompt. Ask the agent to read a file to confirm tool calls work. The request appears in your dashboard usage analytics.

Troubleshooting#

The model is listed but every request fails with 401 missing_api_key. Zed has no key for the provider. Enter it in the Agent Panel, or check that TOKENS_API_KEY is set in the environment Zed started from. A key added to settings.json is ignored.

401 invalid_api_key. The key is wrong, revoked or rotated. Rotation stops the old secret immediately.

404 model_not_found. name must be the exact Tokens ID, including deepseek/. display_name is only a label.

400 invalid_request mentioning a token parameter. Try toggling max_tokens_parameter.

The agent never edits files. Make sure tools is true and the model supports tool calling.

429 errors. Read Retry-After. Troubleshooting explains each rate limit code.

Request details are in Chat Completions. For a terminal agent alongside Zed, see Crush or OpenCode.

Sources: Zed docs, use your own API access, Zed LLM providers, checked October 2026.