CodePilot

CodePilot is a desktop client for AI agents that connects to many model providers through one interface. It runs on macOS, Windows, and Linux as an Electron shell around a Next.js app, keeps conversations in a local SQLite database, and sends API calls straight from the machine to the chosen provider using the user’s own key.

Providers configured out of the box include Anthropic, OpenRouter, AWS Bedrock, Google Vertex AI, Zhipu GLM, Kimi, Moonshot, MiniMax, Volcengine Ark, Xiaomi MiMo, Aliyun Bailian, Ollama, and LiteLLM, alongside any Anthropic-compatible or OpenAI-compatible endpoint. Models can be changed mid-conversation, and conversations run in Code, Plan, or Ask mode with selectable reasoning effort and per-action permission prompts.

The project has grown past coding into a general agent desktop. An assistant workspace stores persona, profile, rules, and memory files at a chosen directory and adds onboarding and daily check-ins; a remote bridge relays sessions through Telegram, Feishu, Discord, QQ, and WeChat; MCP servers and skills extend the tool surface; and a scheduler runs tasks on cron expressions or intervals. Installing the Claude Code CLI adds direct file editing, terminal commands, and git operations.

Features

• Multi-provider access: more than 17 providers plus custom Anthropic- or OpenAI-compatible endpoints, switchable mid-conversation
• Interaction modes: Code, Plan, and Ask, with low-to-max reasoning effort and a thinking mode
• Session control: pause, resume, rewind to a checkpoint, archive, and run two sessions side by side
• Permission control: default or full access, with per-action approval before files are modified
• MCP and skills: stdio, SSE, and HTTP servers with runtime status monitoring, plus custom, project, and global skills
• Remote bridge: drive sessions from Telegram, Feishu, Discord, QQ, or WeChat with per-channel bot credentials
• Assistant workspace: persona, user, rules, and memory files with onboarding, daily check-ins, and persistent notes
• Generative UI: model-created dashboards, charts, and widgets rendered live in the app
• Media studio: image generation with batch tasks, a gallery, and tagging
• Task scheduler: cron-based and interval scheduling that persists across restarts
• Workspace tooling: project file tree with syntax-highlighted preview and a git panel for status, branches, commits, and worktrees
• Usage analytics: token counts, cost estimates, and daily usage charts