airlock social preview
agent-framework206Apache 2.0

airlock

A privacy-first local coding agent that consults frontier AI models without exposing your source code.

Updated Nov 13, 2026
Platforms
macos, linux
Pricing
free-open-source
Status
active
License
Apache 2.0

What it does

Core capabilities at a glance

  • AI Agents
  • Code Privacy
  • Data Loss Prevention
  • Local AI
  • Local LLM
  • Privacy First
  • Privacy Preserving AI
  • Secure AI

Deep dive

The full breakdown - performance, comparisons, and setup

airlockLink to this section

airlock is an agent framework - A privacy-first local coding agent that consults frontier AI models without exposing your source code.

OverviewLink to this section

A local coding agent that keeps your project private and asks frontier models for help without showing them your code.

Your project stays with a model running on your machine, and that model does the work. When it gets stuck on a hard technical point, Airlock asks a frontier cloud model as an outside consultant. The question is rewritten first as a small, generic example, like the minimal reproducible example an experienced developer would post on Stack Overflow. It carries no file names, identifiers, business details or code from your project. You see the exact text before it leaves.

airlock is open-source, written primarily in TypeScript, with 206 GitHub stars under the Apache 2.0 license. It was last updated on 2026-10-28.

Key capabilitiesLink to this section

From the project's documentation:

  • You stay in control.
  • By default you approve each consult.
  • With --approve-consults jewels, only consults that touch your declared crown jewels ask first.
  • Answers are labelled untrusted, and a refusal from the consultant is final.
  • Local agent with workspace-confined tools and a scrubbed environment
  • Abstracted consults with a deterministic gate, crown jewels, cover settings and approval

How it fits a local-AI stackLink to this section

airlock runs on your own hardware, so pair it with a model and a GPU sized to your needs. Use the VRAM calculator to pick a model that fits your card, and see what you can run for hardware guidance. Related agent frameworks in the directory:

SourcesLink to this section

Stats from GitHub, 2026-11-13.

From the community

What users are saying

“Because there is no airlock in the SpaceX Crew Dragon spacecraft used for the flight, the only way to perform a spacewalk is to let all the air out of the spacecraft, and then open one of the hatches. That requires all four crew members to wear spacesuits.”

“Ouch, didn't see your question. Self-upgrade can be initiated by chatting with the agent, it can be manually called via Actions->Upgrade web ui or if you experienced errors during chatting with the agent (tool errors) you can trigger self-repair from the Runs tabs in agent's page.”

Frequently asked

Quick answers to common questions

What is airlock?

airlock is a agent-framework tool for local AI workloads. A privacy-first local coding agent that consults frontier AI models without exposing your source code.

Is airlock free and open source?

Yes, airlock has 206 GitHub stars and is licensed under Apache 2.0. You can self-host it for free on macos, linux.

What platforms does airlock support?

airlock runs on macos, linux.

What hardware do I need for airlock?

The hardware requirements depend on which models you run. Check our hardware directory for compatible GPUs and systems. airlock has 206 GitHub stars and an active community.

Does airlock support GPU acceleration?

airlock supports GPU acceleration via CUDA, Metal, or Vulkan depending on your platform. For the best performance, pair it with an NVIDIA RTX 4090 or 5090.

What are the best alternatives to airlock?

Popular alternatives include other agent-framework tools in our directory. Browse our full collection at /tool for comparisons, community reviews, and benchmark data to find the right fit for your workflow.

How much does airlock cost?

airlock is free-open-source. It is completely free and open source to self-host.

Pairs well with

Complementary tools, models, and hardware