Other coding agents
Status: not tested by Ambit yet. The agents in the table below should work with Ambit, because they speak a format that the Ambit LLM gateway serves. The settings come from each vendor's documentation, read on 2026-10-14. Nobody has run these agents against Ambit. A vendor can change its settings at any time, so check the vendor link first.
Ambit tested Claude Code, Codex, opencode and Hermes Agent; see Claude Code and Cursor and Codex, opencode and Hermes Agent.
What Ambit serves
The LLM gateway accepts exactly these paths: /v1/chat/completions (OpenAI Chat Completions), /v1/messages (Anthropic Messages), /v1/responses (OpenAI Responses), /v1/embeddings and /v1/models. Any other path is refused with 403. Send the personal key as Authorization: Bearer ambit_... or as x-api-key: ambit_.... Clients send the provider's model name.
In the settings below, <llm> is your LLM address (on the quick start: http://llm.localtest.me:8080) and <key> is a personal key that starts with ambit_. Where an agent has a choice of format, choose OpenAI Chat Completions: it is the format with the most testing in Ambit.
Agents with settings from the vendor
| Agent | Format | Settings (not tested) | Vendor documentation |
|---|---|---|---|
| GitHub Copilot CLI | Chat Completions | Environment variables COPILOT_PROVIDER_BASE_URL=<llm>/v1, COPILOT_PROVIDER_API_KEY=<key>, COPILOT_MODEL=<model>. The provider type is openai by default. The model must support tool calls and streaming. | Use your own model key |
| Cline | Chat Completions | In the settings, choose the API provider "OpenAI Compatible". Base URL <llm>/v1, API key <key>, model ID <model>. | OpenAI Compatible |
| Roo Code | Chat Completions | Same three fields as Cline: base URL <llm>/v1, API key, model ID. Roo Code uses native tool calling only, so the model must support it. | OpenAI Compatible |
| Kilo Code | Chat Completions | Add a custom provider with the base URL <llm>/v1, the API key and, if needed, custom headers. Kilo reads the model list from /v1/models. Choose "OpenAI Compatible", not the Responses or Anthropic option. | OpenAI Compatible |
| Continue | Chat Completions | In config.yaml, a model with provider: openai, model: <model>, apiBase: <llm>/v1, apiKey: <key>. For o-series and gpt-5 model names Continue uses the Responses format by default; set useResponsesApi: false to avoid that. Tab autocomplete uses /v1/completions, which Ambit does not serve, so autocomplete does not work. | OpenAI provider |
| Aider | Chat Completions | Environment variables OPENAI_API_BASE=<llm>/v1 and OPENAI_API_KEY=<key>, and the option --model openai/<model>. | OpenAI-compatible APIs |
| Goose | Chat Completions | Environment variables OPENAI_HOST=<llm host>, OPENAI_BASE_PATH=v1/chat/completions, OPENAI_API_KEY=<key>. Goose can also read a custom provider file in ~/.config/goose/custom_providers/ with engine, base_url, api_key_env and models. | Providers |
| Zed | Chat Completions | In settings.json, language_models.openai_compatible.<name> with api_url: "<llm>/v1" and an available_models list that you write by hand. Zed takes the key from the environment variable <NAME>_API_KEY or from the system keychain. | Use your own API keys |
| Factory Droid | Chat Completions | In ~/.factory/settings.json, an entry in customModels with model, baseUrl: "<llm>/v1", provider: "generic-chat-completion-api" and apiKey. Factory says that only the official Anthropic and OpenAI APIs are fully tested. | Bring your own key |
| Qwen Code | Chat Completions | In ~/.qwen/settings.json, an entry in modelProviders.openai with id, envKey (the name of the variable that holds the key) and baseUrl: "<llm>/v1". The schema differs between versions of the Qwen Code documentation, so compare with the version you installed. | Model providers |
| Crush | Chat Completions | In the Crush config file, a provider with type: "openai-compat", base_url: "<llm>/v1" and api_key. The project says that the JSON file format is being replaced, so check the installed version. | Crush on GitHub |
| pi | Chat Completions | In the pi models file, a provider with baseUrl: "<llm>/v1", api: "openai-completions", apiKey and models. The project has an open report (issue 3168) that custom OpenAI-compatible providers can stop responding; we do not know if it is fixed. | Models |
| Warp | Chat Completions | In the Warp settings, add a custom inference endpoint with the URL <llm>/v1 and the key. It works for local agent runs only, not for cloud agent runs. | Custom inference endpoint |
For the MCP gateway, some of these agents may take a URL and a header. Use <mcp>/mcp with the header Authorization: Bearer <key>, as in Connect Codex, opencode and Hermes Agent. We have not checked the exact setting names of these agents; read the vendor's MCP page.
Agents that should not work, or that we cannot say
| Agent | Status |
|---|---|
| Gemini CLI and Antigravity CLI | Should not work today, and nobody has tested it. They speak the Gemini API (/v1beta/models/...:generateContent), and Ambit serves none of the Gemini API paths. One third-party guide says they send the key as x-goog-api-key; we could not confirm this, and Ambit reads only Authorization and x-api-key. |
| Amp, Windsurf, JetBrains AI Assistant | Unknown. We found no documented setting for a custom endpoint in Amp and Windsurf. JetBrains documents an OpenAI-compatible provider, but we could not confirm which request format and headers it uses. |
| GitHub Copilot in VS Code | Unknown. VS Code has "bring your own key" for models, but we have not confirmed the details. Inline completions never use your own key. |
Related
- Connect Claude Code and Cursor: create a key and use it.
- Personal API keys: scope, end dates and revoking.