Skip to content

opencode.md

Latest commit

 

History

History
123 lines (89 loc) · 4.95 KB

opencode.md

File metadata and controls

123 lines (89 loc) · 4.95 KB

OpenCode

OpenCode supports ACP natively via the acp subcommand — no adapter needed.

OpenCode supports 75+ LLM providers via the AI SDK, making it the most flexible backend for OpenAB. Users bring their own provider — no separate API keys per backend needed.

┌──────────┐  Discord  ┌────────┐ ACP stdio ┌──────────┐   ┌───────────────────┐
│ Discord  │◄────────► │ OpenAB │◄────────► │ OpenCode │──►│  LLM Providers    │
│ Users    │ Gateway   │ (Rust) │ JSON-RPC  │  (ACP)   │   │                   │
└──────────┘           └────────┘           └──────────┘   │ ┌───────────────┐ │
                                                 │         │ │ Ollama Cloud  │ │
                                       opencode.json       │ │ OpenAI        │ │
                                       sets model          │ │ Anthropic     │ │
                                                           │ │ AWS Bedrock   │ │
                                                           │ │ GitHub Copilot│ │
                                                           │ │ Groq          │ │
                                                           │ │ OpenRouter    │ │
                                                           │ │ Ollama (local)│ │
                                                           │ │ 75+ more...   │ │
                                                           │ └───────────────┘ │
                                                           └───────────────────┘

Docker Image

docker build -f Dockerfile.opencode -t openab-opencode:latest .

The image installs opencode-ai globally via npm on node:22-bookworm-slim.

Helm Install

helm install openab openab/openab \
  --set agents.kiro.enabled=false \
  --set agents.opencode.enabled=true \
  --set agents.opencode.command=opencode \
  --set 'agents.opencode.args={acp}' \
  --set agents.opencode.image=ghcr.io/openabdev/openab-opencode:latest \
  --set agents.opencode.discord.botToken="$DISCORD_BOT_TOKEN" \
  --set-string 'agents.opencode.discord.allowedChannels[0]=YOUR_CHANNEL_ID' \
  --set agents.opencode.workingDir=/home/node \
  --set agents.opencode.pool.maxSessions=3

Set agents.kiro.enabled=false to disable the default Kiro agent.

Manual config.toml

[agent]
command = "opencode"
args = ["acp"]
working_dir = "/home/node"

Authentication

kubectl exec -it deployment/openab-opencode -- opencode auth login

Follow the browser OAuth flow, then restart the pod:

kubectl rollout restart deployment/openab-opencode

Providers

OpenCode supports multiple providers. Add any of them via opencode auth login:

  • Ollama Cloud — free tier available, models like gemini-3-flash-preview, qwen3-coder-next, deepseek-v3.2
  • OpenCode Zen / Go — tested and verified models provided by the OpenCode team (e.g. opencode/big-pickle, opencode/gpt-5-nano)
  • OpenAI, Anthropic, AWS Bedrock, GitHub Copilot, Groq, OpenRouter — and 75+ more

To list all available models across configured providers:

kubectl exec deployment/openab-opencode -- opencode models

Example: Ollama Cloud with gemini-3-flash-preview

1. Authenticate Ollama Cloud

kubectl exec -it deployment/openab-opencode -- opencode auth login -p "ollama cloud"

2. Set default model

Create opencode.json in the working directory (/home/node). OpenCode reads it as project-level config:

kubectl exec deployment/openab-opencode -- sh -c \
  'echo "{\"model\": \"ollama-cloud/gemini-3-flash-preview\"}" > /home/node/opencode.json'

This file is on the PVC and persists across restarts.

3. Restart to pick up config

kubectl rollout restart deployment/openab-opencode

4. Verify

kubectl logs deployment/openab-opencode --tail=5
# Should show: discord bot connected

@mention the bot in your Discord channel to start chatting.

Notes

  • Tool authorization: OpenCode handles tool authorization internally and never emits session/request_permission — all tools run without user confirmation, equivalent to --trust-all-tools on other backends.
  • Model selection: Set the default model via opencode.json in the working directory using the provider/model format (e.g. ollama-cloud/gemini-3-flash-preview).
  • Frequent releases: OpenCode releases very frequently (often daily). The pinned version in Dockerfile.opencode should be bumped via a dedicated PR when an update is needed.