From 78bf4dbb8ecf557a197d319bfce1c656b624a347 Mon Sep 17 00:00:00 2001
From: thepagent <thepagent@users.noreply.github.com>
Date: Wed, 15 Apr 2026 20:35:40 -0400
Subject: [PATCH] docs: add provider setup, model selection, and Ollama Cloud
 example to opencode guide

Closes #377
---
 docs/opencode.md | 79 +++++++++++++++++++++++++++++++++++++++++++++---
 1 file changed, 75 insertions(+), 4 deletions(-)

diff --git a/docs/opencode.md b/docs/opencode.md
index 8d452d0e..97c4117c 100644
--- a/docs/opencode.md
+++ b/docs/opencode.md
@@ -2,6 +2,26 @@
 
 OpenCode supports ACP natively via the `acp` subcommand — no adapter needed.
 
+OpenCode supports [75+ LLM providers](https://opencode.ai/docs/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
 
 ```bash
@@ -15,12 +35,14 @@ The image installs `opencode-ai` globally via npm on `node:22-bookworm-slim`.
 ```bash
 helm install openab openab/openab \
   --set agents.kiro.enabled=false \
-  --set agents.opencode.discord.botToken="$DISCORD_BOT_TOKEN" \
-  --set-string 'agents.opencode.discord.allowedChannels[0]=YOUR_CHANNEL_ID' \
-  --set agents.opencode.image=ghcr.io/openabdev/openab-opencode:latest \
+  --set agents.opencode.enabled=true \
   --set agents.opencode.command=opencode \
   --set 'agents.opencode.args={acp}' \
-  --set agents.opencode.workingDir=/home/node
+  --set agents.opencode.image=ghcr.io/openabdev/openab-opencode:latest \
+  --set agents.opencode.discord.botToken="$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.
@@ -46,7 +68,56 @@ 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](https://opencode.ai/docs/providers/)
+
+To list all available models across configured providers:
+
+```bash
+kubectl exec deployment/openab-opencode -- opencode models
+```
+
+## Example: Ollama Cloud with gemini-3-flash-preview
+
+### 1. Authenticate Ollama Cloud
+
+```bash
+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:
+
+```bash
+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
+
+```bash
+kubectl rollout restart deployment/openab-opencode
+```
+
+### 4. Verify
+
+```bash
+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.