Neon Pilot Download

A desktop app for AI agents that keep working

Neon Pilot is a local macOS workspace where you chat with an agent, give it files and tools, let it run background work, and keep the results across sessions.

Use it for code, docs, research, automation, and custom workflows that should become part of your agent environment.

Quick install or download the latest release
curl -fsSL https://raw.githubusercontent.com/patleeman/neon-pilot/master/install.sh | bash -s -- --install-cli --bootstrap
Extension-first

Commands, tools, pages, panels, and workflows live in extensions.

Daemon-backed

Automations, persisted runs, follow-ups, and scheduled work keep moving.

Knowledge-aware

Use files, instruction layers, skills, projects, and synced knowledge as context.

What you can do

Start with a conversation. Grow into a workspace.

1

Ask the agent to work

Start a conversation, attach files or folders, and ask Neon Pilot to inspect code, draft docs, debug problems, or plan work.

2

Let work continue

Use background runs, follow-ups, automations, and deferred resumes when a task should keep moving after one reply.

3

Keep reusable context

Connect knowledge, instruction files, skills, project notes, and previous conversations so future turns have memory.

4

Add capabilities

Install extensions or ask the agent to build one when you need a new tool, page, command, integration, or workflow.

You ask for a capability
PMBuild a workflow that triages failed CI logs
9:41 AM
Neon Pilot turns it into an extension
  • Pick the right extension surface
  • Add command palette actions
  • Register a reusable agent tool
  • Wire results into the transcript
It builds, tests, and wires it in
  • Created: ci-triage extension
  • Added: paste/log attachment flow
  • Built: parser and summary action
  • Tests: tool result rendering
  • Docs: workflow usage
It becomes reusable

CI triage ready

Triage failed CI CommandAnalyze logs from the palette, composer, or agent toolbelt.
Neon Pilot desktop app showing a local workspace, conversation, workbench, and file explorer.

Extension-first

Neon Pilot keeps core small. Product workflows become system or user extensions with manifests, commands, tools, UI, settings, and docs.

Learn more →

Background work

Daemon-backed automations, persisted runs, subagents, scheduled tasks, deferred resumes, and follow-ups keep work durable across restarts.

Learn more →

Knowledge and context

Attach files, folders, generated context, instruction files, skills, and git-backed knowledge so agents can carry history.

Learn more →
Neon Pilot command palette screenshot.
Commands
Neon Pilot extensions settings screenshot.
Extensions
Neon Pilot provider settings screenshot.
Provider settings

Self-extension lifecycle

Ask for a feature. Keep the capability.

Neon Pilot is more than a coding agent in a chat box. When your workflow changes, the capability can become a real extension: command-backed, documented, testable, and available again next time.

Features you can ask it to build
  • “Add a settings page for my team’s defaults.”
  • “Create a release checklist for this repo.”
  • “Build a workflow that triages failed CI logs.”
  • “Turn this recurring process into a reusable skill.”
  • “Add a local tool for cleaning stale branches.”
Runtime capabilities
  • Persisted conversations, branches, and transcripts
  • Daemon-backed automations and scheduled tasks
  • Artifacts, telemetry, and activity surfaces
  • Extension host APIs for tools, commands, and pages

Durable runtime

Give your agent a real workspace.

Conversations, runs, automations, command surfaces, model settings, and extension services all share the same local desktop runtime instead of disappearing when a single response ends.

Context and knowledge

Bring the right memory into the turn.

Use the composer, mentions, attached files, generated context, projects, skills, and git-backed knowledge base sync to make each conversation aware of the work around it.

Model provider settings in Neon Pilot.

Ready to make Neon Pilot your own?

Fully open source under MIT. Available on macOS today.

⌄ Download for macOS