Offline is the feature.

Everything CodAI does — completion, chat, reasoning, repository search — happens on your hardware. 4 GB of RAM is enough — 2 GB works with the smallest models. A GPU is optional. A network connection is not required at all.

Download CodAIRead the docs

Why it works on anything.

100% offline AI

No network

Work anywhere without internet or cellular connectivity. All inference and vector indexing execute on your local hardware.

Air-gapped privacy

Local-only

Your codebase, IP, and keystrokes never leave the machine. Zero harvesting, zero retention, nothing to subpoena.

Runs on 4 GB RAM

2–4 GB RAM

The default model runs on 2 GB-RAM machines; 4 GB covers the 1–2 GB class on entry-level laptops and older desktops.

Zero GPU required

CPU-first

Runs natively on standard consumer CPUs — Intel, AMD, Apple Silicon — using SIMD and AVX2 vector instructions.

Portable, plug & play

USB-ready

Run CodAI from a USB drive or external SSD with no installer and no administrator privileges.

15+ languages

Multi-language

Context-aware completion, infill, and technical chat for Python, TypeScript, Rust, Go, C++, Java, PHP, SQL, Shell, and more.

Local vs cloud, row by row.

CapabilityCodAI (local)Cloud assistants
Internet dependencyAir-gapped, 100% offlineConstant connection required
Code privacyNever uploadedCode sent with every request
Monthly costFree, GPL-3.0$10–$20 / month per seat
RAM footprint0.5–1.4 GB for the modelHeavy editor/agent overhead
LatencyLocal RAM speed300–2,000 ms network round-trip
Data leakage riskNo network path existsSubject to breaches & subpoenas

15+ languages, one runtime.

Context-aware syntax intelligence across backend, frontend, systems, and scripting languages — no per-language cloud plugins.

  • Python
  • JavaScript
  • TypeScript
  • Rust
  • Go
  • C / C++
  • Java
  • PHP
  • SQL
  • Bash / Shell
  • HTML & CSS
  • Kotlin
  • Swift
  • C#
  • Ruby

Under the hood.

GGUF model profiles

Pre-configured profiles for Qwen2.5-Coder-1.5B, DeepSeek-R1-Distill, and Llama-3.2, each sized to a RAM budget.

Local repository indexing

A local vector index of your project structure gives the assistant context without uploading a single file.

Model manager

Pull, switch, and remove GGUF weights per project. Models live on your disk; nothing phones home to fetch them at runtime.

VS Code & CLI bridge

Extensions for VS Code, Neovim, and JetBrains IDEs, plus a terminal CLI — all pointed at 127.0.0.1.

Ready to code 100% offline?

Download CodAIView source on GitHub →