pi-devin-local

English | 简体中文

A Pi package that uses Devin Local models inside Pi.

Pi stays the harness. The Devin CLI owns login and the live model catalog (devin auth, devin models list). This is not an ACP integration and does not use Zed.

Fork of kashyab12/pi-devin (npm pi-devin) with additional fixes and features — see What this fork changes. Do not install both: they register the same devin provider.

Why this exists

pi-devin-auth treated Devin as Cascade cloud chat. Models like Sol High, Opus 5, and Fable 5 then failed with:

This model is only in Devin Local.

Those models are available through the local Devin CLI. This package uses that CLI for auth + catalog, then streams completions into Pi so Pi's tools, sessions, and UI stay in charge.

Requirements

• Pi Coding Agent 0.86+
• A signed-in Devin CLI (devin auth status), or a signed-in Devin Desktop
• Node 22.19+ (required by Pi 0.86)

The CLI binary is resolved in this order:

1. $DEVIN_CLI
2. ~/.local/bin/devin, Homebrew, /usr/local/bin/devin
3. Devin.app's bundled devin binary
4. which devin

On Windows, discovery checks the CLI installer and Devin Desktop locations, then uses where.exe to search PATH. Native executables and .cmd/.bat wrappers are supported, including paths with spaces. DEVIN_CLI takes precedence on every platform.

Install

From npm (listed in the package gallery):

pi install npm:pi-devin-local

From git:

pi install git:github.com/mizorewww/pi-devin

Local checkout:

pi install ~/Developers/pi-devin

Restart Pi or run /reload. The Chinese README is at README_zh.md (named without a dot so npm keeps English as the package page default).

Upstream is npm:pi-devin. Install only one package: both register the same devin provider.

Usage

/login devin
/model devin/swe-2
/model devin/claude-opus-5
/model devin/gpt-5.6-sol

Models keep their family id; thinking levels belong to pi and each level is resolved to the matching Devin variant. For SWE-2:

| pi thinking level | model uid sent | |---|---| | medium | swe-2-medium | | high (default) | swe-2-high | | max | swe-2-max |

Use /thinking or shift+tab to change the level; levels a family does not ship are hidden, and Ctrl+S in /thinking saves the startup default. The same applies to every other family (devin/kimi-k3, devin/grok-4.6, …).

/login devin seeds ~/.local/share/devin/credentials.toml from a Devin Desktop sign-in you already have, and otherwise runs devin auth login.

Thinking: the server streams a summary of the model's reasoning; the full trace stays inside the sealed signature and never leaves the server. Pi keeps that summary together with its signature and replays both on the next request, exactly like the Devin CLI, so the model keeps its own prior reasoning across tool calls and turns.

Commands:

• /devin-status — CLI path, version, auth
• /devin-refresh — reload devin models list --format json

The model catalog is cached for six hours in $XDG_CACHE_HOME/pi-devin/models.json (or ~/.cache/pi-devin/models.json). A fresh cache avoids the CLI call at startup; an older cache remains available while it refreshes in the background. Set PI_OFFLINE=1 to skip automatic catalog refreshes. /devin-refresh still requests a refresh explicitly.

What this is / is not

| This package | Not this package | |---|---| | Pi is the agent | Devin taking over the session | | Devin CLI for auth + catalog | Fake Windsurf OAuth paste flow | | Live CLI families (Opus 5, Fable 5, Sol, …) | Hardcoded 11-model cloud allowlist | | Completions streamed into Pi tools | An editor host for Devin |

What this fork changes

Everything upstream does, plus:

• Reuses a Devin Desktop sign-in. Desktop keeps its token in the Electron state DB, so the CLI store stayed empty and /login devin opened a browser for an account that was already signed in. The store is now seeded from windsurfAuthStatus when it is missing.
• One model per family, thinking levels via Pi. devin/swe-2 + /thinking max sends swe-2-max; levels a family does not ship are hidden instead of silently falling back to the default variant.
• Thinking round-trips. The server's thinking summary, its sealed signature and the redacted flag are kept on the block and replayed on the next request, like the Devin CLI does, so the model keeps its own prior reasoning.
• Request shape aligned with the Devin CLI. System prompt in the server's system slot, matching sampling configuration, trajectory reference and planner mode, no stray execution_id.
• Pi payload hooks. onPayload receives the chat request before protobuf encoding: system, messages, tools, modelUid and maxOutputTokens, without authentication or transport metadata. Image parts use Pi's data/mimeType shape. Async inspection, in-place edits and returned replacement payloads are supported, so before_provider_request extensions can observe real requests.

0.3.0 compatibility and validation

Requires Pi 0.86+ and Node 22.19+. Single-variant models now also use the CLI family slug; reselect the model if a saved selection used a raw MODEL_* or variant ID. A cold offline start has no models until a CLI catalog has been cached; run /login devin or /devin-refresh while online. No hardcoded model list is used.

The current CLI price format, tool-result images/error flags, concurrent tool streams, JWT cancellation and session isolation are handled. Incomplete tool JSON and server errors fail explicitly. The stream has a five-minute idle timeout (overridable with timeoutMs), and requests respect the model's output limit. Cost estimates use the CLI's reported rates; an omitted cache rate is unknown and displayed as zero.

Verified on 2026-09-29 against Devin CLI 3000.10.27 / Desktop 3.10.27: SWE-2, GPT-6 Sol and Claude Opus 5.5 completed two-turn tool probes; SWE-2 replayed signed thinking, and Claude read a tool-result image. This private protocol can change. See the audit report for findings and coverage limits, including unverified Gemini-specific thought signatures. On systems without Devin.app, DEVIN_CLIENT_VERSION can override the fallback client version.

Publish

bun run typecheck
npm publish --access public

This is a standard Pi package (keywords: ["pi-package"] + pi.extensions). Once it is on npm with that keyword it is picked up by the package gallery within minutes — there is no separate submission step, and pi has no official namespace for third-party extensions. If it does not show up, bump the version and publish again to force re-indexing.

License

MIT. Unofficial. Not affiliated with Cognition.