123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420

Open-source · AI-native · v0.2.7, signed & notarized

The open-source AI IDE you actually own.

PaddleBoard is an open-source, AI-native IDE — a local-first harness for coding agents. Tell your agent who to be with a PERSONA.md, bring your own model keys, and manage agents, skills, and MCP servers from one place. No telemetry, no hosted plan.

Download Read the docs

macOS (Apple silicon, signed & notarized) · Linux x86_64 tarball · GPL-3.0

PaddleBoard editing a FastAPI project in the PaddleBoard Dark theme

persona

Tell your agent who to be

A PERSONA.md in your repo sets the voice, the constraints, and the house style — for the built-in agent, for subagents, and for Scion. Compose them with extends:. No prompt copy-paste between projects.

fig. 1 · AI Dock — Personas: discovered from the repo, plus a starter library

models

Bring your own keys — or no keys at all

Anthropic, OpenAI, Google, Vertex, Bedrock, or any OpenAI-compatible endpoint. Or run nothing but local: PaddleBoard ships a managed llama.cpp and Gemma weights, so a fresh install can talk to a model without an account.

fig. 2 · Every provider in one list — including Local Models

dock

Every agent surface in one place

Agents, skills, MCP servers, personas, and token usage — the AI Dock is the one window where you install, configure, and watch all of it. No scattered config files.

fig. 3 · AI Dock — Agents: Claude, Codex, Copilot, Cursor, Antigravity

sandbox

Let it run, safely

Agent-run code executes in a microVM, never on your filesystem. The built-in tier runs one-shot commands with nothing to install; Podman unlocks long-running services and sandboxed MCP. Every command still passes your approve/deny gate.

fig. 4 · Sandbox Backend — built-in microVM active

set sail

Ship it from the editor

Set Sail deploys to six serverless platforms using open skill packs from s8sskills.com, with a status-bar indicator that tells you where the build is.

fig. 5 · Set Sail — six platforms, quick deploy or a full pipeline

scion · add-on

Give each agent its own worktree

Scion is a separate tool you install yourself — PaddleBoard detects it and adds a launcher. Each agent gets its own git worktree and container, so several can work in parallel without colliding, each with its own persona.

fig. 6 · Start Scion Agent — task, template, persona

Also in the box

100% Open Source

Every line is public — the editor, the docs, even the deploy playbooks. Fork it, audit it, own it. No hosted plan, no black boxes.

Bring your own keys

OpenAI, Anthropic, Google Gemini, AWS Bedrock, and more. No hosted plan, no upsells — you own the relationship with your model provider.

Private by default

Telemetry is hard-disabled at the source — events never queue, never log, never leave your machine. Even spelling and grammar checking runs fully offline. No account required.

AI Dock

Install and manage agents, skills, personas, and MCP servers from one place — with live logs and an orchestration panel for parallel work.

Personas

Skills say what your agent can do — a PERSONA.md says who it should be. Describe a Senior Dev, an SRE, or a QA engineer and every thread holds that identity, with any model. Or just ask: "be my QA tester."

Agent workbench

A first-class agent panel with delegation, plus support for ADK, LangGraph, CrewAI, and AutoGen agent frameworks out of the box.

Agent Sandbox

Agent-run code executes in ephemeral Podman containers under the gVisor kernel sandbox — your project mounts in, your host stays out. Every command still passes your approve/deny gate.

Set Sail

Serverless-first deploys: ship straight to Cloud Run, AWS Lambda, or Vercel in one click — the agent follows the open-source s8sskills playbook and hands you the helm for auth.

Manifest

The ship's manifest for your repo: repositories, branches, commits, stashes, and contributors in one dockable tree. Click a commit for its diff or a stash to apply it — instant, local, no waiting on git.

Ready from first launch

Opens in PaddleBoard's own theme, then walks you to a working agent before you leave setup — run a model locally with no API key, or bring your own. A short guided tour shows you around.

Built on Zed

All of Zed's speed and multibuffer editing, inherited and kept in sync — with PaddleBoard's AI-native additions layered on top.

An open-source AI IDE

PaddleBoard is an open-source AI IDE: a fast, local-first code editor — built on Zed — with an agentic, AI-native workflow built in. The full source is on GitHub under the GPL. No account, no hosted plan, no telemetry.

A harness for coding agents

Beyond editing, PaddleBoard is a harness for AI coding agents. Bring your own model keys — OpenAI, Anthropic, Google, Gemini Enterprise, Bedrock, Z.ai, Fireworks, and more — run a built-in agent workbench with delegation, and install or orchestrate agents, skills, and MCP servers from the AI Dock, complete with live logs and parallel runs.

Private by default, and yours

Telemetry is hard-disabled at the source, and even spelling and grammar checking runs fully offline. You own the relationship with your model providers, and your code never leaves your machine unless you send it there.

No telemetry. No hosted plan. No account.

PaddleBoard is GPL-3.0 and a fork of Zed. Your keys stay yours, your code stays local, and nothing phones home — the telemetry path is inert by design, not by setting.

Download Get the source