Opencode Go

[filmico]

Hi There,
I'm using the Continue extension for VsCode. As model provider I have the OpenCode Go suscription.

All that models are working fine for chat, edit, etc but not for autcomplete.

I've tried with Ollama and qwen2.5-coder:1.5b-base and the aoutcomplete works fine.

Was fighting with this for 2 days with no luck.

This is the summary provided by the iA of the problem.

The Continue autocomplete function doesn’t work with OpenCode Go providers because their API only exposes /v1/chat/completions and the modern /v1/completions endpoints.
Continue requires either /v1/responses, the legacy OpenAI /v1/completions (2020 format), or an Ollama‑style /api/generate endpoint for predictive autocomplete.
Since OpenCode Go doesn’t implement any of these, Continue cannot stream token‑by‑token predictions.
Chat, edit, and generate work correctly, but autocomplete cannot initialize.

Can Continue add support for OpenAI‑style /v1/completions (non‑legacy) or provide a fallback mode for providers like OpenCode Go?

This is my actual YAML config.

name: OpenCode Go Full Config
version: 1.0.0
schema: v1

models:
  # This is not working
  - name: "DeepSeek V4 Flash Autocomplete"
    provider: openai
    model: deepseek-v4-flash
    apiBase: "https://opencode.ai/zen/go/v1"
    apiKey: "${{ secrets.OPENCODE_GO_API_KEY }}"
    useLegacyCompletionsEndpoint: false
    roles:
      - autocomplete 

  # This is working fine 
  # - name: Qwen2.5-Coder 1.5B Base
  #   provider: ollama
  #   model: qwen2.5-coder:1.5b-base
  #   roles:
  #     - autocomplete


  # --- SOTA Models (State of the Art) ---
  # These are the most powerful models, ideal for software architecture and complex logic.
  - name: "GLM-5.1"
    provider: openai
    model: "glm-5.1"
    apiBase: "https://opencode.ai/zen/go/v1"
    apiKey: "${{ secrets.OPENCODE_GO_API_KEY }}"

  - name: "GLM-5"
    provider: openai
    model: "glm-5"
    apiBase: "https://opencode.ai/zen/go/v1"
    apiKey: "${{ secrets.OPENCODE_GO_API_KEY }}"

  - name: "Kimi K2.6"
    provider: openai
    model: "kimi-k2.6"
    apiBase: "https://opencode.ai/zen/go/v1"
    apiKey: "${{ secrets.OPENCODE_GO_API_KEY }}"

  - name: "DeepSeek V4 Pro"
    provider: openai
    model: "deepseek-v4-pro"
    apiBase: "https://opencode.ai/zen/go/v1"
    apiKey: "${{ secrets.OPENCODE_GO_API_KEY }}"

  - name: "MiMo-V2.5-Pro"
    provider: openai
    model: "mimo-v2.5-pro"
    apiBase: "https://opencode.ai/zen/go/v1"
    apiKey: "${{ secrets.OPENCODE_GO_API_KEY }}"

  # --- BALANCED MODELS ---
  # Great balance between speed and reasoning. Perfect for day-to-day use.
  - name: "Kimi K2.5"
    provider: openai
    model: "kimi-k2.5"
    apiBase: "https://opencode.ai/zen/go/v1"
    apiKey: "${{ secrets.OPENCODE_GO_API_KEY }}"

  - name: "MiMo-V2.5"
    provider: openai
    model: "mimo-v2.5"
    apiBase: "https://opencode.ai/zen/go/v1"
    apiKey: "${{ secrets.OPENCODE_GO_API_KEY }}"

  - name: "MiniMax M2.7"
    provider: openai
    model: "minimax-m2.7"
    apiBase: "https://opencode.ai/zen/go/v1"
    apiKey: "${{ secrets.OPENCODE_GO_API_KEY }}"

  - name: "Qwen 3.6 Plus"
    provider: openai
    model: "qwen3.6-plus"
    apiBase: "https://opencode.ai/zen/go/v1"
    apiKey: "${{ secrets.OPENCODE_GO_API_KEY }}"

  # --- BASIC / FAST MODELS ---
  # The fastest models in the subscription. Ideal for simple refactors or autocomplete.
  - name: "DeepSeek V4 Flash"
    provider: openai
    model: "deepseek-v4-flash"
    apiBase: "https://opencode.ai/zen/go/v1"
    apiKey: "${{ secrets.OPENCODE_GO_API_KEY }}"

  - name: "MiniMax M2.5"
    provider: openai
    model: "minimax-m2.5"
    apiBase: "https://opencode.ai/zen/go/v1"
    apiKey: "${{ secrets.OPENCODE_GO_API_KEY }}"

  - name: "Qwen 3.5 Plus"
    provider: openai
    model: "qwen3.5-plus"
    apiBase: "https://opencode.ai/zen/go/v1"
    apiKey: "${{ secrets.OPENCODE_GO_API_KEY }}"

[filmico]

An update on this...

I was able to configure the autcomplete in Continue by using the Mistral provider with the codestral-latest model and using the free plan of Mistral.

By now, this can be a solution but, it's a pitty at the same time coz I'm paying for the Opencode Go suscription with no luck with the autcomplete on the side of Opencode Go.

At the Docs they have this url available for all the models https://opencode.ai/zen/go/v1/chat/completions

Ref

Any idea on how to solve this?

Here you can find my working config with mistral

name: Discovery Product Workflow
version: 1.0.0
schema: v1

models:
  - name: "Codestral - Autocomplete rápido y consistente para edición diaria"
    provider: mistral
    model: codestral-latest
    apiKey: ${{ secrets.MISTRAL_API_KEY }}
    roles:
      - autocomplete
    defaultCompletionOptions:
      maxTokens: 64
      temperature: 0.1

[at-the-prime]

Just a question why you wouldn't rather use Opencode and use continuedev. is harness any better in continuedev, or is it just a personal choice.

[filmico]

Hi There! It's just a personal need as I combine both ways of coding.

Some times I interact with hands-on the code by also reviewing things, asking questions and refactoring stuffs in the normal way. This is were the autocomplete function it's a must have.

At other times, I launch opencode to work in a more agentic style. The autcomplete have no sense on that way of using this.

My idea was to have inside Vscode a close experience to the one with Github Copilot but, avoiding that suscription.
On top of that, opencode by itself it's a wonderfull tool as well as the continue extension.

[filmico]

@at-the-prime, I forgot to mention... from my point of view, the hasness comes from a well defined AGENTS.md at the root of the project that also points to other agents or skills.
The landscape of providers and models is changing really fast! I think that a good harness system should go with you and your project and be totally independent from IDEs, providers and models.

Thanks for asking ;)