ZenCoder

AI coding assistant — Best things in life are almost free :P

Intelligent model routing · Agent mode · BYOK · Streaming

Overview

ZenCoder is an AI coding assistant that connects to the local ZenCoder daemon (zencoderd) and routes your requests to the best available model — local models first, cloud models when the task demands it.

No subscription. No lock-in. Your keys, your costs.

Features

Feature	Description
🤖 Auto-Routing	Dispatches each request to the optimal model (local vs. cloud)
🛠 Agent Mode	Reads files, proposes multi-file edits, iterates in your workspace
💬 Ask Mode	Single-turn Q&A — fast and focused
📋 Plan Mode	Generate a step-by-step implementation plan
📚 Skills	Auto-detects task type: code review, test gen, refactor, debug
⚡ Streaming	Token-by-token streaming across all backends
🔗 Context	Attach files/folders to include as codebase context
📊 Metrics	Token usage, latency, and iteration count per response

Requirements

ZenCoder requires the zencoderd daemon running locally. Install via the one-line installer:

# Requires GitHub CLI — authenticate first:
gh auth login

# Then install (auto-detects OS/chip, starts zencoderd as a background service):
curl -fsSL https://raw.githubusercontent.com/divyabairavarasu/zencoder-releases/main/install.sh | bash

With at least one model available (Ollama, LM Studio, or a configured cloud key).

Getting Started

Start zencoderd (see above)
Open the Command Palette (Ctrl+Shift+P / Cmd+Shift+P)
Run ZenCoder: Chat
Select a model from the dropdown
Choose a mode: Agent / Ask / Plan
Type your request and press Enter

Commands

Command	Description
`ZenCoder: Chat`	Open the chat panel
`ZenCoder: Explain Selection`	Explain selected code
`ZenCoder: New Chat Session`	Start a fresh session
`ZenCoder: Export Chat Session`	Save session as Markdown
`ZenCoder: Clear Chat`	Clear the current session
`ZenCoder: Detect Local Models`	Refresh the model list
`ZenCoder: Check Service Status`	Verify daemon connectivity

Keyboard Shortcuts

Shortcut	Action
`Enter`	Send message
`Shift+Enter`	New line in input
`Ctrl+U` / `Cmd+U`	Clear chat

Chat Modes

Agent — Full agentic loop. ZenCoder reads workspace files, proposes edits, and iterates. Requires an open workspace folder.
Ask — Standard Q&A. Fast and focused.
Plan — Generates a structured implementation plan without writing code.

Routing Modes

Auto — ZenCoder's router picks local vs. cloud based on the task.
Local — Only use locally-running models (Ollama, LM Studio).
Cloud — Route to cloud providers only.

Settings

Setting	Default	Description
`zenCoder.agentAddress`	`http://127.0.0.1:7777`	Daemon address
`zenCoder.maxMessages`	`100`	Chat history retained per session
`zenCoder.maxContextBytes`	`200000`	Max bytes of file context per request
`zenCoder.contextExtensions`	`.ts .tsx .js .go .py ...`	Extensions scanned for folder context
`zenCoder.contextExcludeFolders`	`node_modules dist .git`	Folders excluded from context scan
`zenCoder.includeWorkspaceByDefault`	`true`	Auto-include current workspace as context
`zenCoder.autoDetectSkills`	`true`	Auto-apply AI skills based on conversation
`zenCoder.lastModel`	`""`	Persists last selected model per workspace
`zenCoder.favoriteModels`	`[]`	Pinned models for quick access

Chat Sessions

Sessions are stored under .zencoder/chat/ in your workspace.

New Session — Start fresh; previous messages are preserved in history.
Export Session — Writes a .md file to .zencoder/chat/exports/.

Development

cd vscode
npm install
npm run build
# Press F5 in VSCode to launch Extension Development Host

Run tests:

npm test           # Unit tests
npm run test:e2e   # End-to-end tests (requires daemon)

Privacy

ZenCoder sends your messages and file context to the model you select. With local models (Ollama, LM Studio), nothing leaves your machine. With cloud models, your data is subject to the respective provider's terms. ZenCoder itself does not collect or transmit any data.

Made with 🧘 — "Best things in life are almost free :P"

ZenCoder AI

Divya Bairavarasu