Use with OpenCode, Cline, Roo Code

Last updated: 24 September 2026

Goalfinder is an OpenAI-compatible gateway, so any coding agent that speaks that format works by pointing its base URL at us. You pay from prepaid credit — no monthly seat, no five-hour rolling window. Create a key in the dashboard to get started with $1 of free credit, no card required, then pick your client below.

OpenCode

Put the config in ~/.config/opencode/opencode.json for every project, or an opencode.json in a project root to share it with a repo. JSON or JSONC both parse.

opencode.json

{
  "$schema": "https://opencode.ai/config.json",
  "model": "goalfinder/deepseek-v4.1-flash",
  "small_model": "goalfinder/glm-5.3-flash",
  "providers": {
    "goalfinder": {
      "name": "Goalfinder",
      "env": ["GOALFINDER_API_KEY"],
      "package": "@opencode/ai/providers/openai-compatible",
      "settings": {
        "baseURL": "https://www.goalfinder.space/api/v1"
      },
      "models": {
        "deepseek-v4.1-flash": { "name": "DeepSeek v4.1 Flash" },
        "glm-5.3-flash": { "name": "GLM 5.3 Flash" },
        "qwen3.8-flash": { "name": "Qwen 3.8 Flash" }
      }
    }
  }
}

Custom providers are an allowlist. Only the ids declared under models show up in /models. OpenCode does not auto-discover GET https://www.goalfinder.space/api/v1/models, so a model missing from this map is invisible to OpenCode.
The package matters. Use @opencode/ai/providers/openai-compatible. The plain openai package speaks the Responses API, which we do not implement.
After saving: export GOALFINDER_API_KEY, restart OpenCode, run /models, and pick a goalfinder/… entry. Keep the key in the environment variable rather than in a committed opencode.json.

terminal

export GOALFINDER_API_KEY=gf-YOUR_KEY

If your OpenCode still uses provider (singular)

Older versions — and most tutorials online — use a different shape. If the block above fails to load, try this one instead. Do not put providers and provider in the same file.

opencode.json — fallback

{
  "$schema": "https://opencode.ai/config.json",
  "model": "goalfinder/deepseek-v4.1-flash",
  "provider": {
    "goalfinder": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "Goalfinder",
      "options": {
        "baseURL": "https://www.goalfinder.space/api/v1",
        "apiKey": "{env:GOALFINDER_API_KEY}"
      },
      "models": {
        "deepseek-v4.1-flash": { "name": "DeepSeek v4.1 Flash" }
      }
    }
  }
}

Cline (VS Code)

Open the Cline panel, then the gear icon → API Provider.
Choose OpenAI Compatible — not OpenAI. The plain OpenAI option pins its own endpoints.
Base URL: https://www.goalfinder.space/api/v1 — include /v1, no trailing slash. Cline appends /chat/completions itself.
API Key: the gf-… key from your dashboard.
Model ID: the exact catalog id, for example deepseek-v4.1-flash. Display names do not work here.
Save, then let Cline verify the connection.

Set Cline's context window and max output from the figures on the model matrix — 128K context and 8K max output for the current flash catalog. We accept image input on no models, so leave vision capabilities off.

Continue (existing installs only)

Continue is no longer maintained. Cursor acquired Continue in June 2026. The project shipped a final v2.0.0 release, the GitHub repository is read-only, and the hosted Hub has shut down — there are no further updates, security patches, or provider-API fixes. The config below still works for a pinned local install, but if you are starting fresh, use Cline or OpenCode instead — both are actively developed and use the same OpenAI-compatible fields shown here.

Edit ~/.continue/config.yaml on macOS and Linux, or %USERPROFILE%\.continue\config.yaml on Windows. It is YAML; config.json is deprecated.

config.yaml

name: Goalfinder
version: 0.0.1
schema: v1
models:
  - name: DeepSeek v4.1 Flash
    provider: openai
    model: deepseek-v4.1-flash
    apiBase: https://www.goalfinder.space/api/v1
    apiKey: ${{ secrets.GOALFINDER_API_KEY }}
    roles: [chat, edit, apply]
    capabilities: [tool_use]
  - name: GLM 5.3 Flash
    provider: openai
    model: glm-5.3-flash
    apiBase: https://www.goalfinder.space/api/v1
    apiKey: ${{ secrets.GOALFINDER_API_KEY }}
    roles: [chat, edit, apply]
    capabilities: [tool_use]
  - name: Qwen 3.8 Flash
    provider: openai
    model: qwen3.8-flash
    apiBase: https://www.goalfinder.space/api/v1
    apiKey: ${{ secrets.GOALFINDER_API_KEY }}
    roles: [chat, edit, apply]
    capabilities: [tool_use]

provider must be openai — Continue's built-in compatible adapter. There is no goalfinder provider, and anthropic will not work.

roles is limited to [chat, edit, apply]. Do not add embed or autocomplete: we have no embeddings endpoint and no fill-in-the-middle support. capabilities: [tool_use] is what allows Agent mode.

If your Continue build does not resolve ${{ secrets.*}} in this file, paste the key literally instead — and keep the file out of version control:

config.yaml

name: Goalfinder
version: 0.0.1
schema: v1
models:
  - name: DeepSeek v4.1 Flash
    provider: openai
    model: deepseek-v4.1-flash
    apiBase: https://www.goalfinder.space/api/v1
    apiKey: gf-YOUR_KEY
    roles: [chat, edit, apply]
    capabilities: [tool_use]
  - name: GLM 5.3 Flash
    provider: openai
    model: glm-5.3-flash
    apiBase: https://www.goalfinder.space/api/v1
    apiKey: gf-YOUR_KEY
    roles: [chat, edit, apply]
    capabilities: [tool_use]
  - name: Qwen 3.8 Flash
    provider: openai
    model: qwen3.8-flash
    apiBase: https://www.goalfinder.space/api/v1
    apiKey: gf-YOUR_KEY
    roles: [chat, edit, apply]
    capabilities: [tool_use]

Reload Continue and pick the model from the dropdown.

Other OpenAI-compatible clients

Roo, Kilo Code, Aider, Hermes, and Cursor's OpenAI compatible provider mode all take the same three fields. Roo Code and Kilo Code are the closest replacements for a Continue install, since they use the same VS Code extension model and are still actively maintained:

settings

base URL:  https://www.goalfinder.space/api/v1
api key:   gf-YOUR_KEY
model:     deepseek-v4.1-flash   # or any id from GET https://www.goalfinder.space/api/v1/models

For a raw SDK call, the Python example in the API reference works unchanged — only the base URL and key differ.

These will not work

Claude Code pointed at ANTHROPIC_BASE_URL. It speaks the Anthropic Messages API, not OpenAI Chat Completions.
Codex and Responses-API clients. We implement Chat Completions only; there is no /v1/responses and no stateful conversation object.
Embeddings or vision. No vector endpoints, and no image input on any model.

Pitfalls

Symptom	Cause	Fix
404 on chat/completions	Base URL missing /api/v1, has an extra /chat/completions, or uses www.	Use https://www.goalfinder.space/api/v1 exactly.
401	Bad or revoked key, or the key is in the wrong field.	Dashboard → create a new key. In Cline, use the OpenAI-Compatible key field.
402	Balance will not cover the worst-case max_tokens.	Buy a pack, or lower max tokens in the client.
OpenCode shows an empty model list	Custom providers are an allowlist.	Declare every id you want under the provider's models map.
Continue talks to api.openai.com	provider is not openai, or apiBase is missing.	Set both: provider: openai and apiBase.
An agent loop burns credit fast	Each tool-call round is billed as a normal completion.	Keep tool_choice and iteration caps tight. See the billing note on tool calling.

The full protocol reference lives at /docs. For data handling — we do not store your prompts, though upstream providers may train on what they receive — see the privacy policy.