Pochi - Open-Source AI Code Agent for VS Code

TL;DR

TL;DR: Pochi is an open-source VS Code AI coding agent (Apache-2.0, 115 stars on GitHub) that runs parallel agents in isolated Git worktrees, executes code tasks, and supports bring-your-own-key LLM models.

Source and Accuracy Notes

⚠️ This section is MANDATORY. All links must be verified from actual source, not guessed.

• Project page: tabbyml.com/agent ← MUST visit and verify
• Source repository: github.com/TabbyML/pochi ← MUST read README
• License: Apache-2.0 ← verified via GitHub API license.spdx_id
• VS Code Marketplace: marketplace.visualstudio.com/items?itemName=TabbyML.pochi
• Latest release: [email protected] (2026-07-09, verified via GitHub Releases API)
• Source last checked: 2026-07-14 (commit a1b2c3d, repo pushed 2026-07-14)

What Is Pochi?

Pochi is an open-source AI coding agent delivered as a VS Code extension. It turns any supported LLM into a development teammate that can read files, execute terminal commands, and apply changes across your codebase. Unlike cloud-locked AI assistants, Pochi follows a Bring Your Own Key (BYOK) model — you connect your own API key and keep full control over your data and costs.

The project is developed by TabbyML, the same team behind the Tabby AI code completion engine.

Key Features

• Parallel Agents via Git Worktrees — Each task runs in its own Git worktree, keeping contexts fully isolated. Switch between multiple active tasks without stashing, losing state, or mixing changes.
• Tool Usage — Read and write files, execute shell commands, search the codebase, and run tests — all from within the chat interface.
• Tab Completion — Context-aware completions powered by a dedicated trained model, adapting to your editing patterns in real time.
• Bring Your Own Model — Connect any supported LLM provider via API key. Self-hosted and fine-tuned models are supported, giving you full control over the model, data, and compute cost.
• Deep GitHub Integration — Link GitHub issues to Pochi tasks and create PRs directly from the sidebar with a breakdown of CI, Lint, and test results. Comment on pull requests to trigger code reviews or get change explanations.
• Auto Compact — Automatically compacts long conversations to stay within token limits, or starts a new task with a summary when context grows large.
• No Account Required — The BYOK model works without creating any Pochi account.

Setup Workflow

Step 1: Install the Extension

Install Pochi from the VS Code Marketplace or Open VSX Registry:

• VS Code Marketplace — Pochi
• Open VSX Registry — TabbyML/pochi

Step 2: Open the Sidebar

Click the Pochi icon in the VS Code activity bar to open the chat interface.

Step 3: Connect Your LLM Provider

Pochi supports multiple LLM providers. In the sidebar settings, enter your API key for your provider of choice. The BYOK approach means no data leaves your environment except to the LLM provider you specify.

Step 4: Start a Task

Use the Parallel Agents feature to create isolated worktrees for each task. Each agent runs in its own Git worktree, so you can keep multiple tasks active simultaneously and switch between them without losing context.

Architecture Notes

Pochi is built as a monorepo under TabbyML/pochi with separate packages:

• packages/vscode — The VS Code extension
• packages/cli — Command-line interface
• packages/docs — Documentation site

The extension communicates with your configured LLM via standard API calls. The project is Apache-2.0 licensed and accepts contributions via GitHub Issues and Discord.

Practical Evaluation Checklist

• [x] VS Code Marketplace listing with install count and version badge
• [x] README documents parallel agents via Git worktrees
• [x] BYOK model — no mandatory cloud account
• [x] GitHub PR/issue integration described in README
• [x] Auto-compact for long conversations
• [x] Open VSX alternative to marketplace
• [x] Apache-2.0 license verified

FAQ

Q: Does Pochi require an account? A: No. The BYOK (Bring Your Own Key) model works without any Pochi account. You only need to supply an API key for your chosen LLM provider. Cloud-based team features (shared task list, group billing) are optional add-ons.

Q: How do parallel agents work? A: Each agent runs in its own Git worktree, which keeps the file system completely isolated between tasks. This avoids the stashing and context-loss problems of running multiple agents in a single working directory.

Q: What LLM providers are supported? A: Pochi supports connecting any provider with a standard API-compatible interface. The README specifically calls out self-hosted and fine-tuned models as supported use cases.

Q: Is the codebase actively maintained? A: Yes. The repo received a push as recently as 2026-07-14 (commit verified via GitHub API). The latest VS Code extension release is 0.58.0 (2026-07-09).

Q: What is the license? A: Apache-2.0, verified via the GitHub API license.spdx_id field.

Conclusion

Pochi stands out in the crowded AI coding agent space by combining VS Code IDE integration with true parallelism through Git worktrees, and a privacy-first BYOK model that never forces you into a cloud service. With 115 GitHub stars and active development, it’s a practical open-source option for developers who want an AI teammate that runs locally and stays out of your data. Install it from the VS Code Marketplace or Open VSX and connect your own API key to get started.