OmniRoute/@omniroute/opencode-plugin

@omniroute/opencode-plugin

Recommended way to use OmniRoute with OpenCode. Pulls a live model catalog from /v1/models (including -low/-medium/-high/-thinking variants as first-class IDs), aggregates combos via /api/combos using a least-common-denominator capability/limit join, sanitizes Gemini tool schemas in flight, and supports multiple side-by-side OmniRoute instances out of the box.

Why this and not @omniroute/opencode-provider?

@omniroute/opencode-provider is the legacy config-generator package — it writes a frozen provider.omniroute block into opencode.json with a hardcoded list of 8 models (OMNIROUTE_DEFAULT_OPENCODE_MODELS). It works on the CLI but in the OpenCode Desktop / Web builds (Tauri / Electron) the runtime re-runs the model picker and the static block surfaces only a few of those — and they drift behind the live OmniRoute catalog.

This plugin solves that by:

• Fetching /v1/models and /api/combos at OpenCode startup, in Node.js — no CORS, no WebView restrictions
• Emitting the provider block dynamically in the plugin's config/provider hook — so opencode.json only needs the plugin entry, not a static provider.omniroute
• Re-fetching on a configurable TTL (default 5 min) and background auto-discovery while OpenCode is running (autoSyncIntervalMs, default 5 min), so new models / combo changes appear without restarting OpenCode
• Exposing a force-refresh path (omniroute_sync_models tool + /omni-sync command template) equivalent to Pi /omni sync
• Computing limit.context for combos as min(member.context_length) from the live catalog (no more null values that cause 4K-token truncation)
• Auto-pickup of interleaved capability for thinking models (merged via PR #3138)

If you only have the legacy opencode-provider block in your opencode.json, replace it with a single plugin entry. No other config changes required — the same auth.json API key works.

Install

The plugin ships pre-built inside the omniroute npm package since v3.8.23. If you have OmniRoute installed, the plugin is already on disk:

# 1. One command — copy the plugin into OpenCode and update opencode.json
omniroute setup opencode --auth

# 2. Follow the interactive prompt to enter your OmniRoute API key
# 3. Restart OpenCode — /models lists the full live catalog

The --auth flag runs opencode auth login --provider opencode-omniroute automatically. Use --base-url to point at a non-default OmniRoute address:

omniroute setup opencode --base-url https://or.example.com --auth

What it does

1. Locates the bundled plugin inside the omniroute installation
2. Copies dist/ + package.json to ~/.config/opencode/plugins/omniroute/
3. Writes/updates opencode.json with the plugin entry (idempotent, replaces legacy entries)
4. (With --auth) runs opencode auth login so the API key is stored

Re-run any time to update the plugin or change the base URL. Older entries for @omniroute/opencode-provider or the legacy opencode-omniroute-auth package are automatically cleaned up.

Manual install (without omniroute CLI)

If you cannot run omniroute setup opencode (local dev, CI, air-gapped), reference the built artifact directly:

cd @omniroute/opencode-plugin && npm run build && npm pack
# then extract into ~/.config/opencode/plugins/omniroute-opencode-plugin/

And add the entry to opencode.json manually (see Quick Start below).

Peer dep: @opencode-ai/plugin (managed by your OpenCode install).

Quick start (single instance, manual)

// opencode.json
{
  "$schema": "https://opencode.ai/config.json",
  "plugin": [
    [
      "./plugins/omniroute-opencode-plugin/dist/index.js",
      {
        "providerId": "omniroute",
        "baseURL": "https://or.example.com",
        // Background re-discovery while OpenCode is running (Pi parity).
        // Default 300000 (5 min). Minimum 60000. Set 0 to disable.
        "autoSyncIntervalMs": 300000,
      },
    ],
  ],
}

opencode auth login --provider opencode-omniroute
# prompts for the OmniRoute API key, writes to ~/.local/share/opencode/auth.json

⚠ Use the --provider flag explicitly. opencode auth login omniroute is parsed as a positional url argument by current OC releases (≤1.15.5) and fails with fetch() URL is invalid. Tracked upstream.

Restart OpenCode. /models lists the full live catalog. Variants (-low, -medium, -high, -thinking) and combos appear as first-class IDs — OmniRoute is the source of truth, no client-side synthesis.

Live catalog refresh (auto + force)

While OpenCode is running, the plugin keeps the model catalog fresh in two ways:

Mechanism	Default	What it does
modelCacheTtl	300000 (5 min)	On-demand TTL: next provider/models hook after expiry re-fetches /v1/models
autoSyncIntervalMs	300000 (5 min)	Background timer: proactively invalidates + re-fetches while the harness is running. Min 60000. Set 0 to disable background polling (TTL still applies)

Force sync now (Pi /omni sync equivalent) — OpenCode has no Pi-style slash-command registration API, so the plugin wires both a tool and command templates:

1. Tool: omniroute_sync_models — invalidates in-memory + disk caches, re-fetches GET /v1/models (and combos/enrichment when enabled), returns { ok, count, ... }.
2. Command templates (type these in OpenCode):
   • /omni-sync — asks the agent to call omniroute_sync_models and report the result
   • /omni-autosync — asks the agent to report current autoSyncIntervalMs / modelCacheTtl status

/omni-sync
/omni-autosync

Multi-instance (prod + preprod side-by-side)

⚠ OC ≤1.15.5 dedupes plugin loads by absolute module path. Two plugin: entries pointing at the same dist/index.js collapse into one (last-listed options win). Workaround: install the plugin twice into separate directories so each entry resolves to a distinct module file. v0.2.x will introduce an instances: [...] shape that registers N providers from a single load.

Dual-install workaround (works today on OC ≤1.15.5)

Pack the plugin once, extract it twice into named directories, then point each plugin: entry at its own copy:

# 1. Build + pack the plugin (run from the plugin worktree)
cd /path/to/OmniRoute/@omniroute/opencode-plugin
npm run build
npm pack
# produces omniroute-opencode-plugin-0.1.0.tgz

# 2. Extract one copy per OmniRoute endpoint
mkdir -p ~/.config/opencode/plugins/omniroute-opencode-plugin-prod
mkdir -p ~/.config/opencode/plugins/omniroute-opencode-plugin-preprod
tar -xzf omniroute-opencode-plugin-0.1.0.tgz -C ~/.config/opencode/plugins/omniroute-opencode-plugin-prod    --strip-components=1
tar -xzf omniroute-opencode-plugin-0.1.0.tgz -C ~/.config/opencode/plugins/omniroute-opencode-plugin-preprod --strip-components=1

Then in ~/.config/opencode/opencode.json reference each directory by absolute path:

{
  "$schema": "https://opencode.ai/config.json",
  "plugin": [
    [
      "./plugins/omniroute-opencode-plugin-prod/dist/index.js",
      {
        "providerId": "omniroute",
        "displayName": "OmniRoute",
        "baseURL": "https://or.example.com",
      },
    ],
    [
      "./plugins/omniroute-opencode-plugin-preprod/dist/index.js",
      {
        "providerId": "omniroute-preprod",
        "displayName": "OmniRoute Preprod",
        "baseURL": "https://or-preprod.example.com",
      },
    ],
  ],
}

Paths are relative to ~/.config/opencode/. Each entry now resolves to a distinct module file, so OC loads them as two separate plugin instances. Authenticate each:

opencode auth login --provider opencode-omniroute
opencode auth login --provider opencode-omniroute-preprod

Each entry gets its own provider id, its own model picker entry, its own slot in auth.json, and its own TTL cache. Closures are isolated per plugin instance — no cross-talk.

After publish (@omniroute/opencode-plugin npm)

Once the package is published, the dual-install becomes two npm install --prefix commands instead of tar -xzf:

mkdir -p ~/.config/opencode/plugins/omniroute-opencode-plugin-prod
mkdir -p ~/.config/opencode/plugins/omniroute-opencode-plugin-preprod
npm install --prefix ~/.config/opencode/plugins/omniroute-opencode-plugin-prod    @omniroute/opencode-plugin
npm install --prefix ~/.config/opencode/plugins/omniroute-opencode-plugin-preprod @omniroute/opencode-plugin

opencode.json paths become ./plugins/omniroute-opencode-plugin-prod/node_modules/@omniroute/opencode-plugin/dist/index.js (and the preprod equivalent).

Features

Feature	What it does	Hook
Dynamic /v1/models	Pulls live catalog (455+ entries on prod) on each refresh, TTL-cached	provider.models
Variants pass-through	-low/-medium/-high/-thinking ship as first-class IDs from OmniRoute (no client synthesis)	provider.models
Combo LCD aggregation	Combos appear with intersected capabilities + min context/output across members	provider.models + config
combo/<slug> namespace + Combo: prefix	Combos surface under combo/claude-primary (not the upstream UUID) and the picker shows Combo: claude-primary so they stand apart from raw provider/model pairs	both hooks
Nice names + cost	/api/pricing/models display names AND /api/pricing per-million-token cost overlaid onto the live catalog	both hooks
Canonical-twin dedup + alias-fallback	/v1/models exposes the same upstream model under both short alias (cc/claude-opus-4-7) and canonical name (claude/claude-opus-4-7); the plugin drops the canonical twin when an alias twin exists (no duplicate rows in the picker) and reverse-maps canonical → alias to pick up enrichment for short aliases (dg/nova-3 → Deepgram - Nova 3) that /api/pricing/models only indexes by canonical	both hooks
Compression pipeline tags	Combo names get tagged with their compression pipeline (e.g. Combo: claude-primary [rtk🟡 → caveman🟠]) when features.compressionMetadata: true. Intensity tokens render as a traffic-light emoji: 🟢 lite/minimal · 🟡 standard · 🟠 aggressive/full · 🔴 ultra	both hooks
Provider-tag prefix	Prepend short upstream-provider label to enriched names (e.g. Claude - Claude Opus 4.7 vs Kiro - Claude Opus 4.7, GHM - GPT 5) so same-id models routed via different upstream connections group visibly in the picker (default-on, opt-out via features.providerTag: false)	both hooks
Usable-only filter	Filter to providers with at least one healthy connection in /api/providers (opt-in via features.usableOnly)	both hooks
Model allowlist/blocklist	Curate the model picker to a fixed set of IDs via features.visibleModels (allowlist) and/or features.hiddenModels (blocklist). Bare suffixes like claude-opus-4-7 match any {prefix}/claude-opus-4-7. Both compose with usableOnly (all filters AND together). Blocklist wins over allowlist (deny takes precedence)	both hooks
Disk-cache fallback	Last-known-good catalog persisted to disk; hydrates on a cold start when /v1/models is unreachable (default-on, opt-out via features.diskCache: false)	config
Bearer injection + suffix-spoof guard	Adds Authorization on baseURL-matched requests only	auth.loader.fetch
Gemini schema sanitization	Strips $schema/$ref/additionalProperties for gemini-*/google-vertex-gemini/*	auth.loader.fetch wrap
Multi-instance	Each plugin entry binds to its own providerId; closures isolated	factory
Config-hook shim	OC ≤1.15.5 fallback: writes static catalog into config.provider[id] (config hook is the only one that fires in serve mode on these versions)	config

Plugin options

Option	Type	Default	Description
providerId	string	"omniroute"	OpenCode provider id; must be unique across plugin entries
displayName	string	"OmniRoute" or OmniRoute (<id>)	Label in the OC UI
modelCacheTtl	number	300000 (5 min)	/v1/models TTL in ms
baseURL	string	resolved from auth.json after /connect	Override OmniRoute base URL
managementReadToken	string	falls back to apiKey	Optional read-only token for management catalog GETs; /v1 inference stays on the connected apiKey
features	object	see below	Feature toggles (all opt-in/out, defaults preserve v0.1.0)

For least-privilege deployments, set top-level managementReadToken to a read-only management token. It is sent only to catalog reads (/api/combos, /api/combos/auto, /api/pricing/models, /api/pricing, /api/context/combos, and /api/providers). Inference requests under /v1, including chat, continue to use the apiKey stored by OpenCode. features.mcpToken remains independent. If managementReadToken is omitted, catalog reads retain the previous apiKey behavior.

features block

Every field is optional. Defaults mirror v0.1.0 behaviour so existing opencode.json files do not need to change.

Feature	Type	Default	What it does
combos	boolean	true	Discover /api/combos and surface them as pseudo-models with LCD capabilities. Combos are keyed under the combo/<slug> namespace and labelled Combo: <name> in the model picker so they're distinguishable from raw provider/model pairs.
enrichment	boolean	true	Pull display names from /api/pricing/models AND per-million-token pricing (input, output, cached → cacheRead, cache_creation → cacheWrite) from /api/pricing, then overlay both onto the live catalog (so the UI shows Claude 4.7 Opus with cost.input: 5, cost.output: 25 instead of raw IDs and zeroed cost).
compressionMetadata	boolean	false	Pull /api/context/combos so combo names get tagged with their compression pipeline, e.g. Combo: claude-primary [rtk🟡 → caveman🟠]. Intensity tokens render as traffic-light emoji (🟢 lite/minimal · 🟡 standard · 🟠 aggressive/full · 🔴 ultra) so the picker advertises "how compressed" each combo is at a glance.
providerTag	boolean	true	Prepend a short upstream-provider label to the enriched display name with " - " separator, so cc/claude-opus-4-7 → Claude - Claude Opus 4.7 differs visibly from kr/claude-opus-4-7 → Kiro - Claude Opus 4.7 in the OC TUI model picker. Label resolution: use /api/pricing/models[<alias>].name verbatim when ≤8 chars (e.g. Claude, Kiro, Codex, Qwen), otherwise fall back to UPPER(alias) (e.g. GitHub Models → GHM, Gemini → GEMINI). Idempotent. Combos intentionally skipped (the Combo: prefix already conveys multi-upstream).
usableOnly	boolean	false	Read /api/providers and filter the catalog to providers that have at least one connection with isActive: true AND testStatus: 'active'. Subtract-filter semantics: providers unknown to BOTH the pricing-models catalog AND the connection table pass through (so synthetic prefixes like agentrouter/* survive). On fetch failure the filter is disabled for the refresh — never hides the whole catalog.
visibleModels	string[]	unset	Allowlist — when set and non-empty, only models whose raw /v1/models ID matches are emitted. Bare IDs (no slash, e.g. claude-opus-4-7) match any {prefix}/claude-opus-4-7; full IDs (e.g. cc/claude-opus-4-7) match exactly. Composes with usableOnly and hiddenModels (all filters AND together). Unset or empty = no filter.
hiddenModels	string[]	unset	Blocklist — models whose raw ID matches are dropped. Same matching rules as visibleModels. When a model is in both visibleModels and hiddenModels, the blocklist wins (deny takes precedence). Composes with usableOnly and visibleModels (all filters AND together). Unset or empty = no filter.
diskCache	boolean	true	Persist the last successful /v1/models + /api/combos + enrichment + connections + compression snapshot to ${OPENCODE_DATA_DIR ?? ~/.local/share/opencode}/plugins/omniroute-<providerId>.json. On a subsequent cold start where /v1/models throws (network down / IP whitelist drop / 5xx) the static block hydrates from the snapshot so OC's model picker survives offline. Soft-fail on read/write — never blocks publishing.
geminiSanitization	boolean	true	Strip $schema/$ref/additionalProperties from tool params when the model id matches gemini
mcpAutoEmit	boolean	false	Auto-write an mcp.<providerId> remote entry into the OC config pointing at <baseURL>/api/mcp/stream with the resolved Bearer token
mcpToken	string	unset	Optional separate Bearer for the auto-emitted MCP entry. Falls back to the provider's apiKey (from auth.json) when unset
fetchInterceptor	boolean	true	Inject Authorization: Bearer + default Content-Type on every outbound request targeting baseURL (suffix-spoof guarded)

Example — enrichment + compression tags + MCP auto-emit

{
  "plugin": [
    [
      "@omniroute/opencode-plugin",
      {
        "providerId": "omniroute",
        "baseURL": "https://or.example.com",
        "managementReadToken": "<read-only-management-token>",
        "features": {
          "combos": true,
          "enrichment": true,
          "compressionMetadata": true,
          "mcpAutoEmit": true,
        },
      },
    ],
  ],
}

With mcpAutoEmit: true, the plugin synthesises an mcp.omniroute entry equivalent to a manual:

"mcp": {
  "omniroute": {
    "type": "remote",
    "url": "https://or.example.com/api/mcp/stream",
    "enabled": true,
    "headers": { "Authorization": "Bearer <apiKey-from-auth.json>" }
  }
}

If you want a narrower-scoped Bearer for MCP (different from the chat/inference key), set features.mcpToken. Operator overrides win: if you already set mcp.omniroute in opencode.json, the plugin will not overwrite it.

Example — production-leaning defaults (clean picker, offline resilience)

{
  "plugin": [
    [
      "@omniroute/opencode-plugin",
      {
        "providerId": "omniroute",
        "baseURL": "https://or.example.com",
        "features": {
          "combos": true,
          "enrichment": true,
          "compressionMetadata": true,
          "usableOnly": true,
          "diskCache": true,
        },
      },
    ],
  ],
}

• usableOnly: true drops models whose canonical provider has no healthy connection in your OmniRoute instance — your /models picker stays focused on what you can actually call.
• diskCache: true (default) writes a snapshot to ${OPENCODE_DATA_DIR}/plugins/omniroute-<providerId>.json on every healthy refresh. On a cold start where /v1/models is unreachable (laptop offline, IP whitelist drop), the snapshot hydrates the static block so OC still shows the catalog instead of a stub.
• compressionMetadata: true annotates combo display names with their pipeline using traffic-light emoji for intensity (e.g. Combo: claude-primary [rtk🟡 → caveman🟠]) so the picker advertises which compression each combo applies and how heavy it is at a glance. Palette: 🟢 lite/minimal · 🟡 standard · 🟠 aggressive/full · 🔴 ultra. Unknown intensities fall through to raw text ([rtk:custom-thing]) so the plugin never hides a value OmniRoute knows but the plugin doesn't.
• providerTag: true (default) prepends a short upstream-provider label so the picker shows Claude - Claude Opus 4.7 for cc/claude-opus-4-7, Kiro - Claude Opus 4.7 for kr/claude-opus-4-7, and GHM - GPT 5 for ghm/gpt-5 (slot.name GitHub Models > 8 chars → abbreviated). Critical when the same model id is sold through multiple upstream connections with different cost/auth/rate-limit profiles. Set to false to keep the pre-v3.8.3 unsuffixed format.

Example — curating the model picker (allowlist + blocklist)

A typical OmniRoute instance serves 600+ models. The OpenCode TUI/CLI picker becomes unusable when you need to scroll through hundreds of entries to find the ~30 models you actually use. visibleModels and hiddenModels let you curate the picker to a fixed set of model IDs that persists in opencode.json across config resets.

{
  "plugin": [
    [
      "@omniroute/opencode-plugin",
      {
        "providerId": "omniroute",
        "baseURL": "https://or.example.com",
        "features": {
          "combos": true,
          "enrichment": true,
          "usableOnly": true,
          "visibleModels": [
            "claude-opus-4-7",     // bare suffix: matches cc/claude-opus-4-7, kr/claude-opus-4-7, etc.
            "cc/claude-sonnet-4-6", // exact: only the cc/ alias
            "gemini-2.5-pro",
            "gpt-5",
            "o3",
            "o3-pro",
            "o4-mini",
          ],
          "hiddenModels": [
            "o3-mini",  // hide the mini variant even if visibleModels is unset
          ],
        },
      },
    ],
  ],
}

• visibleModels is an allowlist — only models whose raw ID matches are emitted. Bare IDs (no slash) match any provider prefix; full IDs (with slash) match exactly.
• hiddenModels is a blocklist — listed models are dropped. When a model is in both lists, the blocklist wins (deny takes precedence).
• Both compose with usableOnly (all filters AND together: a model must pass usableOnly AND visibleModels AND not be in hiddenModels).
• Unset or empty = no filter (current behavior).

@omniroute/opencode-provider is the existing config-generator package — it writes a frozen provider.<id> block into opencode.json at build time. This plugin is the runtime integration.

	@omniroute/opencode-plugin (this)	@omniroute/opencode-provider
Type	OC plugin	Config generator (CLI/build-time)
Models	Live from /v1/models	Frozen at scaffold
Combos	LCD-aggregated live	None
Gemini sanitize	Yes	N/A
OC UI integration	/connect, /models	None
Multi-instance	Native	Manual

Both can coexist; pick the one that fits your environment.

Requirements

• Node >=22.22.3 (per engines.node); tested on Node 22 and 24.
• OpenCode: verified end-to-end against opencode@1.15.5 with @opencode-ai/plugin@1.15.6.
• OC plugin peer (@opencode-ai/plugin) >=1.14.49 for the full feature set (provider hook surfaces models in /models). On <=1.14.48, the plugin falls back to its config hook, writing a static catalog snapshot into config.provider[id] so models still appear.
• The plugin uses the OC v1 plugin shape (default: { id, server }) — older OC releases that only walk named exports will reject it. Stay on OC ≥1.15.

License

MIT. See LICENSE.