Skip to content
| Marketplace
Sign in
Visual Studio Code>Programming Languages>CodePartner AINew to Visual Studio Code? Get it now.
CodePartner AI

CodePartner AI

AnandShah

|
14 installs
| (0) | Free
A powerful agentic AI co-pilot with inline completions, planning, skills, browser control, and MCP support.
Installation
Launch VS Code Quick Open (Ctrl+P), paste the following command, and press enter.
Copied to clipboard
More Info

CodePartner AI

Version VS Code License

CodePartner Hero

CodePartner AI is a powerful agentic AI coding co-pilot that goes far beyond simple chat. Built with an "agent-first" philosophy, it autonomously plans, researches, edits code safely, runs commands, controls a browser, manages Git, and learns reusable Skills — all while following strict engineering best practices.

It supports 5 LLM providers (OpenAI, Azure, Anthropic, Google, Ollama), inline completions, MCP tools, semantic search, multi-agent collaboration, local async agents, git plugin catalog, CI tools via gh, prompt queue, reliable Stop, Antigravity-style @file tags, and a glassmorphism UI (Chat | Plan | Activity — Timeline, Artifacts, Terminal, Skills, Debug under Activity).


✨ Key Features

🧠 Agentic Capabilities

  • Three Operating Modes:
    • Fast Mode: Direct, concise responses and immediate actions
    • Planning Mode: Generates detailed implementation plans + artifacts; waits for approval before editing
    • Architect Mode: Drafts changes in-memory for review before bulk application
  • Multi-Agent System: Automatically dispatches to specialized sub-agents (researcher, code_expert, tester, writer) running in isolated Git worktrees to enable safe, conflict-free concurrent editing. Also supports Custom repo-scoped agents (see below)
  • Proactive Skill Discovery: Suggests saving repeated workflows as reusable global Skills
  • Timeline & Revert: Full audit trail of every tool action with one-click undo
  • Transactional PatchSet: Multi-file accept/reject per agent turn; 3-way merge when you edited the same files
  • Auto-verify: Focused tests after agent edits (codepartner.autoVerify) without waiting for the model
  • NES: Rename variants, path segments, string literals + Finish Changes

🛠️ Powerful Tools

  • Safe File Operations: Always read_file first, precise hunk-based edits (diffLines + selective apply), create_file, grep_search
  • Approval & Safety: Configurable policies (always-ask / full-auto), local sandbox (sandboxMode: off/soft/strict), secret scanner, prompt injection guard
  • Terminal & Testing: Runs shell commands, auto-detects and runs tests (run_tests), with optional inline assist on failures (new terminalInlineAssist setting — offers CodePartner help with failed command + output)
  • Computer-use Browser: Local Chrome — observe interactive elements, click/type/select/scroll/press, screenshot (Puppeteer)
  • Git Integration: Status, stage, AI-generated commit messages (with Co-author), branches, and GitHub PR creation
  • Web & Knowledge: DuckDuckGo search, document indexing (index_docs + query_knowledge)
  • MCP Support: Connect to external Model Context Protocol servers for custom tools
  • Local async agents (no cloud): run_async_agent / list_async_agents — background jobs in the extension host with progress notifications
  • Git plugin catalog: set codepartner.pluginCatalogRepo, run CodePartner: Sync Plugin Catalog; agents merge with .codepartner/agents/*.md
  • CI/CD tools: list_ci_runs, trigger_ci_workflow, write_ci_workflow, plus create_pull_request (uses your gh CLI — no CodePartner backend)
  • Agent Debug panel: mode, model, tokens, recent tools, system diagnostics
  • Sidebar Terminal + multi-file / workspace NES + code-reference scan
  • Custom Agents: Repo-local agents (.codepartner/agents/*.md) with YAML frontmatter + custom system prompts (see dedicated section)
  • Artifacts & Skills: Persistent global storage (~/.codepartner) of code, plans, markdown, logs, and screenshots

🔍 Smart Context

  • @workspace - Semantic search (TF-IDF by default; upgrade to real embeddings via embeddingProvider setting for dramatically better relevance)
  • @web - Real-time web search + sub-agent researcher
  • @file mentions and automatic context from active file + all open tabs
  • Context compaction (LLM summarization when history >40 messages)
  • Token-aware budgeting and truncation (shared across all sources)
  • Image/file attachments support
  • New in v2.1.3: Real embedding-based @workspace search via embeddings.ts (provider-agnostic for OpenAI/Azure/Google/Ollama, cosine similarity, chunking with overlap, content-hash caching, graceful TF-IDF fallback)

🎨 Premium UX

  • Beautiful glassmorphism sidebar with tabs (Chat, Plan, Timeline, Artifacts, Skills)
  • Separate thinking panel when the model streams reasoning (collapsible Thought block above the answer)
  • Inline ghost text completions (toggle with Ctrl+Shift+I)
  • Diff views, apply/insert/copy actions on code blocks
  • Model selector, mode switcher, chat history with rename/delete
  • Feedback button for bugs/features

📌 What's new in v2.8.0

Safety & runtime

  • Local sandbox (codepartner.sandboxMode: off | soft | strict) — filtered env, command safety, optional Linux bubblewrap
  • Capability approvals, path containment, secret redaction, prompt-injection tracking (web / @file / MCP / browser)
  • Tool scheduler + per-path write locks
  • Reliable Stop, prompt queue, turn-id cancellation

Agent edits

  • Transactional PatchSet — multi-file accept / reject / 3-way merge review after each turn
  • Auto-verify — focused related tests after edits (codepartner.autoVerify: off | focused | full)
  • Review policy packs — security, react, node, quality
  • NES / Finish Changes — renames, case variants, path segments, string literals, multi-file

Context & UX

  • Symbol index + DocumentSymbol enrichment, context ranker, multi-root preferred folder
  • Sidebar Chat | Plan | Activity (Timeline, Artifacts, Terminal, Skills, Debug)
  • Computer-use browser — observe → act → observe (local Chrome)
  • Optional codepartner.completionModel for inline completions
  • Local async agents, git plugin catalog, CI via gh

See CHANGELOG.md for the full 2.4 → 2.8 history.

🚀 Quick Start

  1. Install from the VS Code Marketplace
  2. Configure API (secure storage):
    • Run "CodePartner: Set API Key" from Command Palette (Ctrl+Shift+P). This stores your key securely via VS Code SecretStorage (migrates any old settings.json value automatically).
    • Then set:
      • provider (e.g. openai, anthropic, ollama)
      • model (e.g. gpt-4o, claude-3-5-sonnet-20241022, llama3)
    • Optional: tweak approvalPolicy, sandboxMode, autoVerify, completionModel, contextTokenBudget, inlineCompletions, etc.
  3. Open Sidebar: Click the robot icon in Activity Bar or press Ctrl+L / Cmd+L
  4. Try it:
    • Fast: "Explain this function"
    • Plan: Switch to Planning mode and ask "Add user authentication with JWT"
    • Use slash commands: /fix, /explain, /test

Tip: Use @workspace for project-wide awareness and @web for research.

Lint: npm run lint · auto-fix: npm run lint:fix · types+lint fix: npm run fix


⌨️ Commands & Keybindings

Command Title Keybinding Description
codepartner.focus Focus on Chat Sidebar Ctrl+L / Cmd+L Open the sidebar
codepartner.toggleInlineCompletions Toggle Inline Completions Ctrl+Shift+I / Cmd+Shift+I Enable/disable ghost text
codepartner.explainSelection Explain Selected Code Ctrl+Shift+; / Cmd+Shift+; (when selection) Context-aware explanation
codepartner.fixErrors Fix Errors in File Ctrl+Shift+' / Cmd+Shift+' Auto-fix diagnostics
codepartner.generateTests Generate Tests for File Ctrl+Alt+T / Cmd+Alt+T Create unit tests
codepartner.newChat New Chat Ctrl+Alt+N / Cmd+Alt+N Start a fresh conversation
codepartner.cancelActiveTask Cancel Active Task Ctrl+Alt+X / Cmd+Alt+X Stop any running agent/task
codepartner.acceptPatchSet Accept Patch Set - Keep all files from last agent turn
codepartner.rejectPatchSet Reject Patch Set - Roll back all files from last agent turn
codepartner.reviewPatchSet Review Patch Set - 3-way merge / diffs for pending patch set
codepartner.showDebugLog Show Debug Log - Open output channel
codepartner.setApiKey Set API Key - Securely store LLM API key

Inline Completions: Ctrl+Right (next word), Ctrl+Alt+Right (next line)

Slash Commands in Chat: /fix, /explain, /test, /compact, /clear, /plan, /architect


⚙️ Configuration

Key settings (full list in VS Code Settings → search "codepartner"):

| Setting | Default | Description | |---------|---------|-------------| | codepartner.provider | openai | LLM provider (openai, azure, anthropic, google, ollama) | | codepartner.embeddingProvider | disabled | New in v2.1.3: Upgrade @workspace to real neural embeddings (same-as-chat / openai / azure / google / ollama). Defaults to zero-cost TF-IDF with automatic fallback on errors. | | codepartner.embeddingModel | | Embedding model (blank = sensible default per provider) | | `codepartner.embeddingEndpoint` | | Custom embeddings endpoint (blank = provider default) | | codepartner.apiKey | `` | Set via "CodePartner: Set API Key" command (secure SecretStorage) | | codepartner.model | gpt-4 | Chat model / deployment name | | codepartner.apiEndpoint | https://api.openai.com/v1 | Custom chat endpoint | | codepartner.approvalPolicy | always-ask | Safety (always-ask / ask-for-shell / full-auto / yolo) — controls confirmation for shell, file edits, and git operations | | codepartner.terminalInlineAssist | disabled | New in v2.1.7: When terminal commands fail, optionally offer CodePartner help (disabled / codepartner-only / all-terminals). "all-terminals" watches every terminal (requires explicit permission confirmation) | | codepartner.contextTokenBudget | 6000 | Shared token budget for file context, open tabs, and @file mentions | | codepartner.maxTokens | 4096 | Max output tokens per response | | codepartner.inlineCompletions | false | Enable ghost-text inline completions | | codepartner.inlineCompletionDebounce | 500 | Debounce (ms) for inline suggestions | | codepartner.mcpServers | {} | Custom Model Context Protocol tool servers | | codepartner.addAICoAuthor | true | Add Co-authored-by to Git commits made by CodePartner |

Tip for embeddings: Start with "same-as-chat" (reuses your existing key/provider). Set to disabled to avoid any extra API cost/latency. See embeddings.ts for supported models and graceful fallback behavior.


📁 Global Persistence (~/.codepartner)

  • Skills: Reusable instruction sets (.md files)
  • Artifacts: Generated plans, code snippets, screenshots, logs
  • Knowledge: Indexed documentation via index_docs + query_knowledge
  • Global Instructions: global_instructions.md applied to every chat

Skills and knowledge travel with you across all projects.


🛠️ Custom Repo Agents & Plugin Catalog

Workspace agents live in .codepartner/agents/*.md (YAML frontmatter + body). Optionally set codepartner.pluginCatalogRepo to a git URL and run CodePartner: Sync Plugin Catalog (or the sync_plugin_catalog tool) to pull shared agents from a public/private repo into extension storage — no marketplace backend.

Local async agents

Use run_async_agent to start a background sub-agent (progress toast, cancellable). Results via list_async_agents. Jobs live only while VS Code is open.

CI via GitHub CLI

Tools list_ci_runs, trigger_ci_workflow, write_ci_workflow call your installed gh CLI. create_pull_request remains for PR flow.

🛠️ Custom Repo Agents (format)

Define your own specialized agents local to the current repository by placing Markdown files in a .codepartner/agents/ directory (create it if it doesn't exist — the extension will detect it automatically).

Each file uses YAML frontmatter for metadata + the rest of the file as the full system prompt:

---
name: senior_frontend
description: Expert in React, Tailwind, and modern frontend best practices
tools: read_file, edit_file, grep_search, run_tests, list_dir
---

You are a senior frontend engineer. 
- Always call read_file before any edit_file.
- Provide complete, production-ready implementations with tests.
- Follow the project's existing style and architecture.

Key behaviors:

  • name: is optional (falls back to filename without .md)
  • tools: is a comma-separated list; only valid/known tools are used (unknown tools are logged as potential typos but do not cause failure)
  • Files without proper frontmatter, empty body, or only instructions are skipped
  • list_custom_agents tool returns available agents or a helpful bootstrap message if none found (encouraging you to create the folder + first agent)
  • Use via call_custom_agent tool or by referencing in run_parallel_agents task list
  • Fully unit-tested with edge cases (whitespace, missing fields, invalid tools, graceful degradation)
  • Testing tip: In Extension Development Host (F5 from source), create .codepartner/agents/ inside the opened test workspace and add .md files — they will be loaded on next agent run

Custom agents integrate seamlessly into the agent loop and parallel execution without changing core behavior or safety rules.

See CHANGELOG.md and src/customAgents.ts for parser/implementation details.


📸 Screenshots

(Screenshots will be added to the marketplace listing. Current assets in /media)

  • Glassmorphism Sidebar with multiple tabs
  • Planning Mode with task tracker
  • Timeline showing tool executions and revert options
  • Artifact gallery with screenshots
  • Inline completions in editor
  • Diff review workflow

🔄 How It Works (Agent Rules)

CodePartner follows strict internal guidelines:

  1. Always read_file before edit_file
  2. Use exact text matching for search/replace
  3. Create full, complete implementations (no truncated code)
  4. In Planning Mode: Research → Create Plan Artifact → Wait for "proceed" → Execute → Verify with tests
  5. Explains reasoning in responses
  6. Uses markdown formatting

See full system prompt and tool definitions in src/extension.ts.


🤖 vs Other Tools

CodePartner excels in:

  • Broadest provider support including local Ollama (no subscription required)
  • Global reusable Skills and proactive suggestions
  • Computer-use browser (observe/act loop) and local sandbox
  • Timeline with undo and Architect drafting mode
  • Zero lock-in — bring your own keys

For a detailed comparison (including gaps like inline completions in earlier versions), see codepartner_comparison_analysis.md.

Note: Current release v2.8.0 — local sandbox, PatchSet accept/reject/merge, auto-verify, computer-use browser, richer NES, review packs, multi-root, symbol index, and the earlier Stop/queue/artifacts/file-tags/Agent Debug/async agents/CI stack. See CHANGELOG.md for the full history.


📄 License

This project is licensed under the MIT License.


Developed with ❤️ by AnandShah

GitHub Repository | Report Issue | Marketplace Page


Happy Coding! Try asking CodePartner to "build a full-featured todo app with tests" in Planning mode.

  • Contact us
  • Jobs
  • Privacy
  • Manage cookies
  • Terms of use
  • Trademarks
  • Your Privacy Choices
  • Consumer Health Privacy
© 2026 Microsoft