finds.dev← search

// the find

op7418/CodePilot

★ 6,489 · TypeScript · NOASSERTION · updated Sep 2026

A multi-model AI agent desktop client — connect any AI provider, extend with MCP & skills, control from your phone. Built with Electron + Next.js.

An Electron + Next.js desktop shell that wraps 17+ AI providers (Anthropic, Bedrock, Vertex, several Chinese providers, Ollama, custom endpoints) behind one chat UI, with MCP/skills support and remote control via Telegram/Discord/etc. Aimed at developers who want one client to switch between Claude Code CLI and other model backends without being locked into a single vendor.

Provider abstraction is real, not just a marketing list — you can switch models mid-conversation without losing context, and it covers cloud (Bedrock/Vertex), direct API, and local/self-hosted (Ollama, LiteLLM) without treating any of them as second-class. Session handling goes beyond a chat wrapper: pause/resume, rewind to an arbitrary checkpoint, and split-screen dual sessions are genuine engineering, not UI polish. Storage is local SQLite in WAL mode, so there's no forced cloud dependency, and it can import existing Claude Code CLI .jsonl history instead of starting cold. The release pipeline is unusually disciplined for an Electron project — signed/notarized macOS builds with checksum and attestation gates, and docs that explicitly tell users not to bypass Gatekeeper rather than the usual workaround-first README.

It's BSL-1.1, not permissive open source — commercial use requires a paid license from the maintainer (converts to Apache 2.0 in 2029). That's fine, but worth knowing before you build on it. Windows builds are unsigned (expect SmartScreen prompts), and the in-app updater only checks SHA-512 metadata from the GitHub Release rather than doing real Authenticode verification, which is a weaker trust story than the macOS path. Scope has crept well past 'AI coding client' into image generation, cron-based task scheduling, and bridges to five separate chat platforms (Telegram/Feishu/Discord/QQ/WeChat) — each bridge is its own credential surface to configure and keep secure in an app that's already holding your provider API keys. The docs/exec-plans folder is huge and heavy on post-hoc firefighting — entries like 'windows-runtime-recovery-hardening', 'codex-runtime-crash', 'oom-recovery', and 'electron-port-stability' suggest real production instability has shown up more than once; worth skimming recent issues before relying on it day to day.

View on GitHub → Homepage ↗

// want more like this?

We dig through GitHub every week and send a few repos picked for what you actually care about — each with an honest take like this one.

Get finds in your inbox → Search again →