Autonomous coding, wherever you work

Go Deep
into Code.

An AI coding agent that reads your whole project, runs commands, and writes & verifies code — with any model, hosted or local. From your terminal, Mac, or editor.

npm install -g codeep

See Codeep in action

Free
Open-source CLI (Apache 2.0)
11 providers incl. Ollama
No account

CLIInstallv3.14.0 macOSDownloadv3.10.0 VS CodeInstallv3.0.0 ZedSet upvia ACP

Codeep

Current task

Refactor authentication flow

Running

Understand current flow
Map dependencies
Design refactor approach
Plan changes
Run baseline tests
Ready to edit

src/server/auth/flow.ts+142 −78

118async function authenticate(req: Request) {

119 const user = await getUser(req.headers.authorization);

120 if (!user) {

121 throw new Error('Unauthorized');

122 }

123 return user;

124 const session = await sessionService.verify(req);

125 if (!session.isValid) {

126 throw new AuthError('Unauthorized');

127 }

128 return session.user;

129}

All 48 tests passedNo failures · completed in 3.2s

Resourcesestimated

Electricity2.1–10 Wh

Cooling water0.56–11 mL

Research-based range, not provider telemetry.

Modelgpt-5.6-sol

Context1.05M · 42 files

Tokens24.7K

Cost (est.)$0.14–$0.24

Codeep · Dashboard

Live tour

Agent working

01 · Observe

Every run, visible.

Follow active sessions, token usage, cost, project activity, and estimated resource impact from one focused cockpit.

Live session feed
Cost and token trends
Energy and cooling estimates

Explore the Dashboard

How it works

From prompt to verified code

Codeep handles the full development loop autonomously — no context switching, no copy-pasting.

Describe
Type the task in plain language, no special syntax. Codeep reads your project structure and key files for context.
Plan & execute
It breaks the work into steps and runs them: editing files, running commands, fetching docs. You watch it in real time.
Verify
It runs your build, tests or typecheck, and fixes what breaks before it hands the change back.

codeep fix the login form - validation isn't triggering on submit

Scanning projectReading src/components/LoginForm.tsx
Identifying issueMissing onSubmit handler — e.preventDefault() not called
Applying fixPatching LoginForm.tsx (2 lines)
Verifying✓ Build passed · ✓ No type errors

Fixed. The form handler was missing e.preventDefault(), causing the page to reload before validation ran.

Get Codeep

Up and running in 30 seconds

The same agent in your terminal, on your Mac, and inside VS Code or Zed. Pick the one that fits how you work.

Node.js 20 or later
Any terminal — zsh, bash, fish, PowerShell
An API key for your AI provider

CLI

macOS, Linux, Windows (WSL)

npm install -g codeep

Then run codeep in your project — no configuration
The engine the VS Code and Zed integrations run on
Open source, Apache 2.0

Installation guide

macOS app

Native app · Apple Silicon & Intel

Download for Mac.dmg, notarized by Apple

Built in SwiftUI, not an Electron wrapper
Project sidebar, inline diffs and tool confirmations
Auto-updates via Sparkle

Mac app docs

VS Code

Extension · VS Code Marketplace

Terminal

code --install-extension VladoIvankovic.codeep

# or search “Codeep” in the Extensions view

Open in Marketplace

Chat sidebar with a live agent plan
Diff preview before you click Allow
Needs the Codeep CLI on your PATH

Extension docs

Zed

Agent Client Protocol (ACP)

~/.config/zed/settings.json

"agent_servers": {
  "Codeep": {
    "type": "custom",
    "command": "codeep",
    "args": ["acp"]
  }
}

Pick Codeep in the agent panel, then type /help
All 30+ slash commands, with live tool calls
Needs the Codeep CLI — Zed runs codeep acp

Zed setup guide

AI Providers

Use any LLM you want

Codeep works with all major AI providers. Switch between them at any time from the settings menu. Your keys are stored securely on your machine.

Anthropic

Claude — Fable, Opus, Sonnet and Haiku

Claude Opus 5.5Claude Fable 5.1Claude Fable 5

DeepSeek

Pro and Flash, both with a 1M context window

DeepSeek V4.1 FlashDeepSeek V4 Pro

Google AI

The Gemini line, from Gemini 4 Argon down to the Flash-Lite tier

Gemini 4 ArgonGemini 3.1 ProGemini 3.8 Flash

Grok (xAI)

Agentic coding and flagship models

Grok Build 0.1Grok 4.7Grok 4.6

Kimi (Moonshot)

Coding plan and pay-per-use

Kimi CodeKimi K3Kimi K3 (256K)

MiniMax

Coding plan and pay-per-use

MiniMax M3.1 FlashMiniMax M3

Ollama (Local AI)

Run models locally or on a remote server — no API key required

Any installed modelLlamaQwen Coder

OpenAI

The GPT-5.6 ladder — Sol, Terra and Luna

GPT-6.1 SolGPT-6 AstraGPT-6 Sol

OpenRouter

One key, 100+ models from every major provider

100+ modelsopenrouter/auto

Qwen (Alibaba)

Coding Plan, Token Plan, pay-per-use, and live ModelScope catalog

Qwen3.7-PlusQwen3.6-PlusQwen3.5-Plus

Z.AI

Coding plan and pay-per-use

GLM-5.3GLM-5.3 Flash

Any OpenAI-compatible endpoint

vLLM, LiteLLM, LM Studio or your own server — no API key required

Any served modelCustom Base URL

Switch anytime with /settings. Prompt caching is automatic on Anthropic (and OpenRouter → Anthropic routes): a typical agent session is 60–90% cheaper. How caching works

Capabilities

Everything you need to code with AI

From autonomous tasks to git workflows — in your terminal, Mac app, or editor.

Autonomous agent

Reads files, runs shell commands, edits code, and auto-verifies the result (build / typecheck / tests) — all in one loop. You describe the goal; Codeep handles the steps.

File editsShell execAuto-verify

Sub-agents

Delegate a self-contained sub-task to a specialist that runs in its own fresh context and reports back — researcher, reviewer, tester, or your own. Keeps the main thread focused; optional auto-review after every change.

delegate/agentsAuto-review

Plan mode

Preview the agent's full plan — files it'll touch, commands it'll run, risk it self-assesses — before any code changes. /plan <task> → review → /go to execute.

/plan/goPre-execution review

Approval on your phone

When a run stops on a dangerous tool it waits — and if you have walked away, it waits for as long as you are gone. Connect a Telegram bot and the same question arrives on your phone. Answer in either place; whichever comes first decides. Terminal and Mac app, on every platform.

/telegramRun / Skip / CancelNo server

Full project context

Auto-scans your project structure, detects language and framework, reads key files. Project memory and lifecycle hooks persist across sessions.

Auto-scan/memory.codeep/hooks/

Audit record

Every run writes down what it touched — reads and refusals, not just changes. Shared by the CLI, the Mac app and VS Code. Paths and commands only; never file contents.

/audit.codeep/audit/Refusals

Git workflows

AI review of your diff — bugs, security, performance, edge cases — and semantic commit messages. Code review also runs on every PR with the GitHub Action.

/review/diff/commit

Secure & local

API keys stored locally (encrypted at rest, synced encrypted if you opt in). Dangerous commands require confirmation. Full undo history and /checkpoint snapshots.

Local keys/checkpoint/undo

See all features

Extend

Tools and skills, one command away

Plug in Model Context Protocol servers for your files, the web and your devices, and install skills the agent invokes on its own.

MCP marketplace

Codeep speaks the Model Context Protocol. Curated servers install with /mcp install <id>.

Files & Code
Filesystem
Read, list, and (optionally) write files in a sandbox directory.
/mcp install filesystem
Web
Playwright (browser)
Drive a real browser — navigate, click, fill, screenshot, and scrape — via Microsoft Playwright. The de-facto browser-automation MCP (supersedes Puppeteer).
/mcp install playwright
Files & Code
Git (local repo)
Inspect commits, blame, diffs, and branches in a local git repository.
/mcp install git
Web
Brave Search
Web, local, image, video and news search via the Brave Search API.
/mcp install brave-search
Utility
Memory (knowledge graph)
Persistent knowledge graph the agent reads/writes across sessions.
/mcp install memory
Testing & Devices
iOS Simulator
Drive the iOS Simulator — tap, type, swipe, screenshot, record, install & launch apps. macOS only; needs Xcode + idb (`brew install facebook/fb/idb-companion`).
/mcp install ios-simulator

Browse all 9 servers

Plus /mcp add for any custom server (npx, binary, or HTTP endpoint)

Skill marketplace

Drop a SKILL.md in .codeep/skills/ and the agent auto-invokes it when triggers match. Publish to share, install on any machine with one command. Claude Code-compatible format.

update-dependenciesv1.0.0
0
Safely bump project dependencies for any detected ecosystem (npm/pnpm/yarn/bun, pip, cargo, go, SwiftPM): update within semver-safe ranges or a single named package, reinstall, then run build and tests to verify nothing broke — reverting or flagging on failure and summarizing what changed plus breaking notes.
VladoIvankovicinstalls
/skills install VladoIvankovic/update-dependencies
pr-reviewv1.0.0
0
Review the current uncommitted or branch diff before it ships: pull the diff with git, read changed files for context, and report correctness bugs, missing error handling, security issues, and test gaps grouped by severity, each with file:line and a concrete fix. Use before committing, opening a PR, or when asked to "review my changes / diff / branch." Read-only by default.
VladoIvankovicinstalls
/skills install VladoIvankovic/pr-review
swift-pkg-testv1.0.0
0
Build and test a Swift Package Manager package on macOS: run `swift build` then `swift test`, parse compiler and test failures, fix the source, and re-run until the build is clean and all tests pass. Use when asked to build/test/fix a Swift package, get a SwiftPM target green, or triage failing `swift test` output.
VladoIvankovicinstalls
/skills install VladoIvankovic/swift-pkg-test

Browse marketplace

Or write your first skill — Markdown + YAML frontmatter, ~5 minutes

Releases

Shipping fast

RSS All releases →

FAQ

Common questions

Don't see yours? Start a discussion on GitHub.

Is my code sent to a third party?
Only to the LLM provider you choose, only when you invoke the agent. Codeep is a thin client over your API key — file contents, prompts, and tool calls go straight to Anthropic / OpenAI / Z.AI / OpenRouter / etc. We don't proxy anything through codeep.dev. The optional dashboard sync (opt-in via codeep account) uploads session metadata and learning preferences only — never raw code. Privacy details.
How does it compare to Cursor / Claude Code / Aider?
Codeep is terminal-native (TUI + ACP for Zed / VS Code), with a native Mac app that runs the same agent — built in SwiftUI, not an Electron wrapper. It's multi-provider by design, and its CLI is open-source. Cursor is a fork of VS Code locked to its own backend. Claude Code is Anthropic-only and closed-source. Aider is terminal-native too but doesn't do ACP or skill bundles. If you want one tool that works the same across the CLI, the Mac app, Zed, and VS Code with any provider — Codeep's the bet. Mac app docs.
Does it work offline?
Yes — pick ollama as your provider. Codeep talks to a local Ollama daemon, so no network call leaves your machine. Tool calls (read/write files, run commands) are unchanged. Smaller local models are slower at multi-step reasoning, so set /settings → Max Iterations higher and expect more back-and-forth. Provider docs.
What does it cost?
Codeep is free to use — the CLI is open source (Apache 2.0), and the VS Code extension and GitHub Action are MIT. You pay your LLM provider directly. For OpenRouter, Codeep shows the exact usage.cost from each response on your dashboard. Prompt caching (default on Anthropic) typically reduces a session bill by 60-90%. For Z.AI on the GLM Coding Plan, it's flat-rate. How caching works.
Which model should I use?
For coding: Claude Opus 5 (Anthropic) or GPT-5.6 are the strongest. For cost-per-token: DeepSeek V4 Pro or GLM-5.2 are 5-10× cheaper with little quality drop on typical tasks. For local: Llama 3.1 70B via Ollama (needs 24GB+ VRAM). Or set provider to openrouter and model openrouter/auto — the router picks for you per request. Provider matrix.
Can I use my own LLM endpoint?
Yes — pick Custom (OpenAI-compatible) with /provider, then set /settings → Custom Base URL to your endpoint, including /v1. If it needs a key, add one with /login. Self-hosted vLLM, LiteLLM, LM Studio, or any inference server that speaks OpenAI's /v1/chat/completions works. Custom endpoint docs.
What's MCP and why should I care?
Model Context Protocol — the emerging standard for letting LLMs talk to external systems like a database, GitHub or a browser. Codeep supports it end-to-end. Browse the MCP servers.
Do I need to log in to codeep.dev?
No — Codeep works fully offline with just an LLM key. The optional account (free, GitHub OAuth) syncs session history, profiles, and learning preferences across machines, and unlocks the skill marketplace publish flow. Everything is opt-in. Account docs.

Start coding with AI today

Free to use. Works with your existing AI provider subscriptions. No account required to get started.

npm install -g codeep

Download for Mac

Read the docs

Go Deepinto Code.

Every run, visible.

From prompt to verified code

Describe

Plan & execute

Verify

Up and running in 30 seconds

CLI

macOS app

VS Code

Zed

Use any LLM you want

Anthropic

DeepSeek

Google AI

Grok (xAI)

Kimi (Moonshot)

MiniMax

Ollama (Local AI)

OpenAI

OpenRouter

Qwen (Alibaba)

Z.AI

Any OpenAI-compatible endpoint

Everything you need to code with AI

Autonomous agent

Sub-agents

Plan mode

Approval on your phone

Full project context

Audit record

Git workflows

Secure & local

Tools and skills, one command away

MCP marketplace

Filesystem

Playwright (browser)

Git (local repo)

Brave Search

Memory (knowledge graph)

iOS Simulator

Skill marketplace

Shipping fast

v3.14.0

v3.13.0

v3.12.1

Common questions

Start coding with AI today

Go Deep
into Code.