devin-opencode-provider

Use Devin subscription models from OpenCode and compatible coding agents by speaking Devin's Connect-RPC agent protocol.

This project is a custom AI SDK provider (LanguageModelV3) plus an OpenCode plugin that handles authentication and model discovery. Instead of calling a generic chat-completions API, it encodes and decodes Devin's protobuf agent protocol over HTTP/2 to Devin's agent backend.

Status: Usable end-to-end for authentication, model discovery, streaming, and tools. See Known limitations.

Demo

OpenCode driving a Devin-routed model through this provider:

Features

• OpenCode integration — registers a devin provider with auth hooks and cached model list (classic 1.x plugin, plus a dedicated OpenCode 2.0 entry)
• Authentication — browser OAuth (PKCE), or API key from devin.ai/settings
• Model discovery — fetches available models from Devin's API and caches them locally
• Image input — advertises vision only for supported models and forwards OpenCode image attachments to Devin
• Streaming — bidirectional Connect-RPC chat with token usage tracking
• Tool calls — maps Devin tool messages to AI SDK / OpenCode tool-call parts
• Safe file reads — strips OpenCode 1.x/2.0 read wrappers, preserves truncation meaning, and refuses whole-file writes copied from partial reads
• Thinking / reasoning — surfaces extended-thinking deltas where the model supports it

Requirements

• Bun (for development and tests)
• OpenCode
• An active Devin account with API access

Installation

This package is one plugin with host-specific entrypoints. Use the guide for your OpenCode major version:

Host	Guide
OpenCode 1.x	This section — classic plugin plus optional 1.18 plugin/v2
OpenCode 2.0	docs/opencode-2.md — plugin/opencode2 only

Do not load the 1.x and 2.0 entrypoints in the same host.

OpenCode 1.x (~/.config/opencode/opencode.json):

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

OpenCode 2.0 (prefer a dedicated OPENCODE_CONFIG_DIR):

{
  "plugin": ["devin-opencode-provider/plugin/opencode2"]
}

From npm / bun

npm install devin-opencode-provider
# or: bun add devin-opencode-provider

Pin a version if you want: devin-opencode-provider@0.1.5.

From a local clone

git clone https://github.com/oakimov/devin-opencode-provider.git
cd devin-opencode-provider
bun install
bun run build

Point OpenCode 1.x config at the built files with absolute file:// URLs:

{
  "plugin": ["file:///absolute/path/to/devin-opencode-provider/dist/plugin.js"],
  "provider": {
    "devin": {
      "npm": "file:///absolute/path/to/devin-opencode-provider/dist/index.js",
      "name": "Devin",
      "models": {}
    }
  }
}

For OpenCode 2.0 local development, see docs/opencode-2.md.

OpenCode setup

Host config, auth commands, and local-clone wiring differ by major version — follow OpenCode 2.0 for 2.0-specific details.

Authenticate

Host	How
OpenCode 1.x	opencode auth login → devin
OpenCode 2.0	/connect → Devin (see docs/opencode-2.md)

Then choose a method:

Method	Description
Devin account (browser login)	PKCE OAuth — opens devin.ai to sign in
API key	Paste a key from devin.ai/settings

After login, the plugin fetches your available models and writes them to <host-cache>/devin-models.json. On later startups, a missing, empty, or expired cache is refreshed during config load when Devin auth is available. DEVIN_API_KEY / WINDSURF_API_KEY are also picked up automatically for auth without /connect.

Paths (host cache)

Model/version caches and Devin project metadata live under a host cache root, resolved in this order:

1. Explicit createDevin({ cacheDir }) / host API cache override
2. Native OpenCode: $XDG_CACHE_HOME/opencode, otherwise ~/.cache/opencode

Kind	Default (OpenCode)	Notes
Model and version cache	~/.cache/opencode/
OpenCode auth (auth.json)	~/.local/share/opencode/	$XDG_DATA_HOME/opencode/ when set
OpenCode config (AGENTS, skills, …)	~/.config/opencode/

Select a model

Devin discovery returns many flat wire ids (claude-opus-5-max, swe-1-7-lightning-medium, …). This provider collapses them into one OpenCode model id per family with parameter-only variants. Pick the base id, then a variant:

# Effort ladder (Claude Opus 5): Low, Low Fast, Medium, …, Max, Max Fast
opencode run --model devin/claude-opus-5 --variant Max "Hello from Devin via OpenCode"

# SWE-1.7: Medium, Max, Lightning Medium, Lightning Max
opencode run --model devin/swe-1-7 --variant Max "Refactor this module"
opencode run --model devin/swe-1-7 --variant "Lightning Max" "Quick fix"

# Flat / unsuffixed ids still work when Devin only exposes one member
opencode run --model devin/swe-1-6-slow "Hello from Devin via OpenCode"

In the TUI: choose devin/<base-id>, then open the variant picker.

Variants

Devin models often expose parameterized variants (effort, thinking, fast, context tier, …). The plugin materializes those as OpenCode model variants. In the TUI, pick one from the variant dialog or cycle with OpenCode's variant_cycle keybind (default ctrl+t).

The selected variant's Devin parameter map is forwarded on the request (isolated under providerOptions.devin.devinVariantParameters so unrelated OpenCode options are not leaked onto the wire). The provider validates that explicit selection against the current cached tuple; malformed, reordered, or stale selections fail clearly instead of silently falling back to another variant.

1M / long context

OpenCode's context limit is static per model entry, while Devin's long-context tier is a variant parameter (context=1m). When a model has both a base tier and a 1m tier, the plugin emits a separate OpenCode entry <model-id>-1m (for example claude-opus-4-6-1m) with:

• limit.context set to the 1M window (so overflow checks and compaction match the tier)
• limit.output set to 128000 (max generation tokens — not the context window; base entries use 32000)
• only the long-context variants in its picker
• the real Devin model id carried in options.devinModelId (not config.id, which would make OpenCode merge base variants into the 1M entry)

The request still uses Devin's original model id; OpenCode's synthetic -1m id is only for picking and limits.

Max vs Cursor Max Mode

Max here is a high-effort variant (effort=max), not Cursor's IDE Max Mode toggle (which sets wire max_mode and unlocks long-context tiers). Devin has no equivalent Max Mode flag in this catalog. Longer context is a separate base id when Devin exposes it (for example devin/claude-opus-4-6-1m with Thinking / No Thinking).

OpenCode selection	Variant params	Typical Devin wire id
devin/claude-opus-5 + Max	effort=max	claude-opus-5-max
devin/claude-opus-5 + Max Fast	effort=max, fast=true	claude-opus-5-max-fast
devin/swe-1-7 + Lightning Max	lightning=true, effort=max	swe-1-7-lightning (alias)

Image input

The provider advertises text + image input only when the selected Devin model supports images; model output remains text. This works in the classic plugin, the 1.18 v2 plugin, and OpenCode 2.0.

Image support is resolved from Devin's AvailableModels.supportsImages metadata. OpenCode image file parts are decoded from bytes, base64/data URLs, local file URLs, or HTTP(S) URLs and sent through Devin's message fields. Attachments are limited to 20 MiB total, matching OpenCode's desktop attachment budget. PDF, audio, and video inputs are not advertised or silently discarded.

Programmatic usage

import { createDevin } from "devin-opencode-provider"

const devin = createDevin({
  name: "devin",
  accessToken: process.env.DEVIN_ACCESS_TOKEN,
  apiKey: process.env.DEVIN_API_KEY,
  // apiBaseURL: "https://api.devin.ai",
  // cacheDir: "/path/to/host/cache",
})

const model = devin.languageModel("swe-1-6-slow")
// model implements AI SDK LanguageModelV3 (doStream / doGenerate)

Pass either accessToken (JWT from OAuth or key exchange) or apiKey (raw key). Optional: apiBaseURL, cacheDir, headers, workspaceRoot. You can also set DEVIN_API_KEY as an environment variable instead of passing apiKey.

Environment variables

Variable	Description
DEVIN_API_KEY	Devin API key (sk-ws-01-..., cog_..., or devin-session-token$...) for auth without /connect
WINDSURF_API_KEY	Alias for DEVIN_API_KEY (for Windsurf compatibility)
DEVIN_API_BASE_URL	Override API base URL (default https://api.devin.ai)
DEVIN_PROVIDER_SHOW_DISABLED	Set to 1 / true to include plan-disabled cascade models in the catalog (useful for debugging Pro vs full lists)
DEVIN_PROVIDER_DEBUG	Set to 1 or true to enable wire-level debug logging
DEVIN_PROVIDER_DEBUG_FILE	Override debug log path (default: $TMPDIR/devin-provider-logs-<uid>/debug-<pid>.log)
DEVIN_OPENCODE2_DEV_ENTRY	Local OpenCode 2.0 only. Absolute path to a built entry file (usually dist/index.js). Rewrites the AI SDK package to aisdk:file://… so the daemon imports your local build instead of npm install-ing the published package. Export它 before opencode2 service start, then restart after rebuilds. Unset in production. See OpenCode 2.0 local clone.
DEVIN_OPENCODE2_TODOS	OpenCode 2.0 only. Set to 1 or true to register plugin-owned todowrite/todoread (in-memory; no TUI sidebar). Off by default. OpenCode 1.x still uses the host builtin.
XDG_CACHE_HOME	Base for host cache dirs ($XDG_CACHE_HOME/opencode/)
XDG_DATA_HOME	When set, OpenCode auth.json is read from $XDG_DATA_HOME/opencode/ instead of ~/.local/share/opencode/

Development

bun install          # install dependencies
bun run build        # compile TypeScript → dist/
bun run typecheck    # type-check without emit
bun test             # run unit tests

Architecture

OpenCode
  └── DevinPlugin (auth, model cache, config hook)
        └── createDevin() → LanguageModelV3
              ├── language-model.ts  AI SDK adapter
              ├── protocol/   protobuf messages, framing
              └── auth.ts     PKCE OAuth, API key exchange

Module	Role
src/plugin.ts	OpenCode hooks: provider registration, OAuth, API key exchange
src/plugin-core.ts	Host-neutral SDK factory, API base resolution
src/model-config.ts	Flat Devin models → OpenCode bases + parameter-only variants
src/index.ts	createDevin factory; default export is DevinPlugin
src/language-model.ts	AI SDK LanguageModelV3 adapter (doStream, doGenerate)
src/auth.ts	PKCE OAuth, API key exchange, JWT refresh
src/models.ts	Model discovery and devin-models.json cache
src/protocol/	Protobuf encode/decode, chat framing, usage parsing
src/plugin-opencode2.ts	OpenCode 2.0 plugin ({ id, setup, server })

Package exports

Import path	Export
devin-opencode-provider	createDevin, DevinPlugin (named + default)
devin-opencode-provider/plugin	DevinPlugin (classic OpenCode 1.x Hooks)
devin-opencode-provider/plugin/v2	OpenCode 1.18 Promise v2 plugin
devin-opencode-provider/plugin/opencode2	OpenCode 2.0 plugin ({ id, setup, server })
devin-opencode-provider/server	Same module as plugin/opencode2 (Host.resolve)

The package root intentionally stays plugin-safe for OpenCode's classic loader. OpenCode 2.0: see docs/opencode-2.md.

Troubleshooting

Problem	What to try
No Devin models in the picker (OpenCode 1.x)	Confirm Devin auth (opencode auth login → devin). Restart OpenCode — if auth is present and the cache is empty, models are fetched on startup. Confirm provider.devin.npm is the package name (or a built file://…/dist/index.js).
No Devin models in the picker (OpenCode 2.0)	Confirm /connect → Devin (or shared auth.json). Use a dedicated OPENCODE_CONFIG_DIR (not mixed with 1.x). Ensure $OPENCODE_CONFIG_DIR/plugins/devin/ is a package directory re-exporting plugin/opencode2 (not a bare .js). After auth, Devin models appear in the picker from the in-memory inventory — no opencode.json model list is required. Filter by provider Devin (time.released is not 0, but models may sort lower). See docs/opencode-2.md.
Auth / 401 errors mid-session	Re-login. OAuth and exchanged API-key JWTs refresh automatically when near expiry; a revoked refresh token needs a fresh login.
Local OpenCode 2.0 still runs the published package	Set DEVIN_OPENCODE2_DEV_ENTRY to an absolute …/dist/index.js path before starting the daemon, persist it with opencode2 service set env DEVIN_OPENCODE2_DEV_ENTRY …, rebuild (bun run build), then opencode2 service restart. Loading only dist/plugin-opencode2.js is not enough — without the env var, 2.0 still npm installs the published package into the host cache.
Empty or stale model list / wrong variant order	Fully restart the OpenCode TUI after bun run build (long-lived processes keep the old plugin). Delete <host-cache>/devin-models.json (default ~/.cache/opencode/) only if discovery itself is stale. Existing Devin auth is enough to refill the cache; re-login only if auth itself is broken. Set DEVIN_PROVIDER_SHOW_DISABLED=1 if your plan hides most models.
Need wire-level logs	Set DEVIN_PROVIDER_DEBUG=1 (optional DEVIN_PROVIDER_DEBUG_FILE; the default is debug-<pid>.log under $TMPDIR/devin-provider-logs-<uid>/) and reproduce the issue.

Security

Project instructions may reference absolute or ~/ paths (OpenCode parity). See SECURITY.md for the trust model and OPENCODE_DISABLE_PROJECT_CONFIG.

Known limitations

• Personal use / ToS — this provider speaks Devin's private agent protocol. Use only with an account you own; Devin may change or restrict the API without notice.
• Token usage tracking — usage counts are extracted from Devin's protocol frames (ModelUsageStats / ResponseStatistics) and mapped to AI SDK LanguageModelV3Usage. Cache-read/write metrics are available when Devin provides them.
• Max vs Cursor Max Mode — There is no Cursor-style Max Mode toggle; Max = high effort only. Longer context is a separate base id (-1m), not a Max Mode flag.

License

MIT