Version 3.3 refreshes the built-in presets and adds a private, self-hosted channel. Refreshed agent presets The six starting presets now mirror how a real solo-operator agent fleet is set up today — a founder-side copilot as the main agent, with isolated specialists for operations, research, social, building, and support. Each preset's role, jobs, channels, isolation, and default model are up to date. Buzz — your own private relay Buzz is now a first-class channel: a self-hosted relay you run yourself, so your agent can be reached and a fleet can coordinate without any third-party messaging account. It comes with its own credential card for the relay URL. Inspect before you apply A new Inspect preview shows a preset's full configuration — build mode, focus, jobs, channels, bridges, and model — before you commit to it. Sharper preflight The readiness check now flags when Buzz is approved without a relay URL, so you never export an agent that advertises a channel it can't reach. No account, no analytics, no tracking. Generated files are written only to folders you choose.

# What's New ## Version 3.2 Build 6 · macOS ### Agent Forge Builder identity The app has been reworked around a generic agent-package builder identity. Public-facing app name, copy, screenshots, and documentation now present Agent Forge Builder instead of relying on third-party runtime branding. ### Launch Readiness Brief Preview + Export now includes a pro launch-readiness card with a 0-100 score, readiness tier, capability map, blocker count, warning count, and a copy-ready Markdown brief. Every export includes `LAUNCH_BRIEF.md` with: - Agent identity, slug, score, and launch tier. - Capability map for providers, channels, bridge packs, data sources, founder context, learning loop, and secret handling. - Handoff checklist for blockers, file preview, secret posture, and planned file count. - Open preflight findings with severity and fix guidance. ### Safer handoff workflow The existing pre-export checks now pair with the launch brief so users can see both issue-level validation and an executive handoff summary before writing files. No account, no analytics, no tracking. Generated files are written only to folders the user chooses.

What's New Version 3.1 Build 4 · macOS OpenClaw Field Guide — a full offline handbook, always available A new reference section you can open at any time, even before you finish the wizard. Because this sandboxed edition can't run the OpenClaw installer for you, the Field Guide is how you finish the job — and the reference you keep coming back to once your agent is live: - A copy-ready install + first-run quickstart (check Node, install the CLI, run `openclaw onboard --install-daemon`, confirm the gateway, open the dashboard). - Step-by-step instructions for applying an exported pack into your live `~/.openclaw/workspace`, including backing up first and merging the config patch by hand. - The full OpenClaw CLI cheatsheet, grouped and copyable. - Troubleshooting first-aid for the common "gateway not responding / agent frozen / where do I start" situations. - Reference for workspace files, supported channels, built-in skill categories, model providers, environment variables, and onboarding flags. - A plain-English glossary (gateway, daemon, workspace, heartbeat, channel, skill, bridge, provider, and more). All of it works with no internet connection. Reference content is drawn from the official OpenClaw documentation at docs.openclaw.ai. Pre-export Preflight Check — catch whole-build mistakes before you ship A new validator runs across your entire build (not just one page) right on the Preview & Export screen and flags the real cross-section problems the per-step nudges can miss: - Missing agent name, or no model provider selected (blocking). - Daily questions enabled with no approved channel to ask in. - A BlueBubbles or WhatsApp bridge selected without its matching channel. - Founder-profile injection enabled with an empty or unfed profile. - Real API keys about to be baked into a pack you might share. - Jobs that don't match the chosen primary focus, missing data-source keys, and more. Each finding is graded — Needs fixing, Worth a look, or Heads up — with a one-click "Fix in …" jump straight to the page that owns it. The reference now travels with your pack Every export now includes two new documents: - `FIELD_GUIDE.md` — the same install-to-troubleshoot reference, bundled into your workspace so the knowledge stays with the files. - `BUILD_CHECKS.md` — the exact preflight result captured at export time, as a paper trail you can act on later. Polish - The wizard step counter and progress now count only the real, visible wizard steps, so "Step X of N" is accurate in this edition. No account, no analytics, no tracking. Generated files are written only to folders you pick.