opencode-auto-fallback

OpenCode plugin that automatically detects model errors and switches to a fallback model chain — with intelligent retry backoff for transient failures.

Features

• Two-tier classification: immediate fallback for auth errors, exponential backoff retry for rate limits and transient failures — detected via structured SDK error types, not text matching
• Fallback chain: ordered list of fallback models per agent, with variant/reasoning/temperature support
• Per-model timed cooldown: failed models are skipped until cooldown expires
• Default‑retry safety net: any unrecognized error is treated as retryable
• Zero config startup: auto-generates fallback.json with sensible defaults on first run
• Toast notifications: terminal toasts when fallback is triggered
• Large context fallback: automatically forks the session with a larger context model when context fills up, then injects the result back into the main session with structured context

Installation

1. Register in opencode config

Add "opencode-auto-fallback" to the plugin array in ~/.config/opencode/opencode.json:

{
  "$schema": "https://opencode.ai/config.json",
  "plugin": ["opencode-auto-fallback"],
}

2. Configuration (optional)

On first run, a default config is auto-created at ~/.config/opencode/fallback.json. You can customize it:

{
  "$schema": "https://raw.githubusercontent.com/HyeokjaeLee/opencode-auto-fallback/main/docs/fallback.schema.json",
  "enabled": true,
  "defaultFallback": ["anthropic/claude-opus-4-7"],
  "agentFallbacks": {
    "reviewer": [
      "zai-coding-plan/glm-5.1",
      {
        "model": "openai/gpt-5.5",
        "temperature": 0.5,
        "reasoningEffort": "medium",
      },
    ],
  },
  "cooldownMs": 60000,
  "maxRetries": 2,
  "logging": false,
}

Using with oh-my-openagent

When an agent is provided by oh-my-openagent, the runtime agent name can differ from the key in oh-my-openagent.json. Configure fallback with the name that appears in OpenCode session logs, normalized by removing whitespace and ignoring case.

For example, if logs show agent=Sisyphus - Ultraworker, configure it as Sisyphus - Ultraworker (whitespace is stripped, case is ignored):

{
  "agentFallbacks": {
    "Sisyphus - Ultraworker": [
      {
        "model": "opencode-go/deepseek-v4-pro",
        "variant": "high",
      },
    ],
    "hephaestus - deepagent": ["zai-coding-plan/glm-5.1"],
  },
  "largeContextFallback": {
    "agents": ["Sisyphus - Ultraworker", "hephaestus - deepagent"],
    "model": "opencode-go/deepseek-v4-pro",
  },
}

Agent matching is exact after normalization: Sisyphus - Ultraworker and sisyphus - ultraworker match the same entry, but sisyphus does not automatically match Sisyphus - Ultraworker.

Auto Updates

The plugin checks for updates on every startup and installs them automatically — no manual intervention needed.

opencode starts → check npm registry → newer version? → bun/npm update → done

If the auto-update fails for any reason, a toast notification appears with the manual update command.

Large Context Fallback

When an agent's context window fills up mid-task, automatically fork the session with a larger context model to finish the work without losing the main session's progress. The main session waits while the fork handles the work, then receives the result with structured context about what happened during compaction.

{
  "largeContextFallback": {
    "agents": ["sisyphus", "explore"],
    "model": "openai/gpt-5.5",
    // Optional: minimum ratio difference required to trigger fallback
    "minContextRatio": 0.1,
  },
}

The plugin reads context window sizes from the SDK's model metadata automatically (input.model.limit.context). When both the current model and the large fallback model have been used in the session, their limits are known and the 10% ratio check is applied. If the large model hasn't been used yet (first compact), its limit is unknown and the fallback proceeds without the ratio check.

Flow:

original model working → context full → auto compact triggered
    → plugin forks the session with the large context model
    → main session pauses (auto-continue suppressed)
    → forked session processes the full context with large model
    → forked session completes → result injected into main session
        with compaction context:
        • conversation was compacted (context was full)
        • the last task before compaction
        • the fork's result
    → main session continues with original model and full context

Note: Manual /compact commands do not trigger large context fallback — only automatic compaction (when context fills up from an assistant response) activates it.

Behavior Details

• Session Forking: When compaction triggers, the plugin forks the current session. The forked session uses the large context model while preserving the full conversation history.
• Main Session Pause: The main session's auto-continue is suppressed during fork execution. It idles until the fork completes.
• Structured Result Injection: When the fork finishes, the result is injected into the main session with:
  • A notice that context was compacted
  • The last user request before compaction (as context reminder)
  • The fork's response
  • An explicit instruction to continue work based on the result
• Fork Timeout: If the forked session doesn't complete within a configurable timeout, it's marked as failed and the main session proceeds with normal compaction.
• Cooldown Safety: If the large context model is in cooldown (e.g., from a previous error), the fallback is skipped and normal compaction proceeds.

Fallback Model Entry

Each entry in a fallback chain can be a simple string or an object:

// Simple
"openai/gpt-5.5"

// With options
{
  "model": "openai/gpt-5.5",
  "variant": "high",
  "temperature": 0.5,
  "reasoningEffort": "medium",
  "maxTokens": 8192
}

How It Works

Error Classification

The plugin detects errors through session.error events (structured statusCode and isRetryable flags) and session.status events (message-based pattern matching for rate limits and transient errors).

Retry Flow

With the default maxRetries: 2:

1st failure → abort → wait 2s   → re-prompt with SAME model
2nd failure → abort → wait 4s   → re-prompt with SAME model
3rd failure → FALLBACK CHAIN: try next model in ordered list

Immediate fallback errors (quota, auth) skip retries entirely and go straight to the fallback chain.

Fallback Chain

The plugin tries each model in the chain sequentially. Models in cooldown are automatically skipped. If all models are exhausted, the error is logged and a critical toast is shown.

Compatibility with Other Fallback Plugins

If another plugin with model fallback logic is installed alongside this one, place opencode-auto-fallback first in the plugin array. The first plugin in the list processes the model response first — by placing this plugin first, it intercepts the error before other fallback plugins see it.

// ✅ opencode-auto-fallback handles errors first
"plugin": ["opencode-auto-fallback", "other-fallback-plugin"]

// ❌ Other plugin may interfere
"plugin": ["other-fallback-plugin", "opencode-auto-fallback"]

Development

# Install dependencies
bun install

# Type check
tsc --noEmit

# Run tests (82 tests)
bun vitest run

# Bump version (CI auto-publishes)
npm version patch --no-git-tag-version
git push