ForkScape v1.0.0

ForkScape is an AI workspace built on a branching canvas instead of a chat scroll.
Every question is a node; every answer is its child. Because each call sends only its
own branch — not the whole transcript — deep, exploratory work stays cheap and focused.
It runs entirely on your machine, with local models or your own API keys.

---

Download

Platform	File	Notes
Windows	ForkScape_1.0.0_x64-setup.exe	Standard installer
Windows	ForkScape_1.0.0_x64_en-US.msi	For managed/IT deployment
macOS	ForkScape_1.0.0_universal.dmg	Intel + Apple Silicon
Linux	ForkScape_1.0.0_amd64.AppImage	Any distro — chmod +x then run
Linux	ForkScape_1.0.0_amd64.deb	Ubuntu / Debian
Linux	ForkScape-1.0.0-1.x86_64.rpm	Fedora / RHEL

What's in it

Branch instead of scroll — fork any answer into parallel follow-ups that never
contaminate each other. The Thread panel reads any single branch as a normal chat.

Bring your history — import ChatGPT, Claude, and Gemini exports, plus Claude Code
and Codex CLI session logs. An organizer agent reads a flat conversation and rebuilds it
as a topic tree, titling every node.

Research Mode — ask once, several models answer in parallel, and a judge model merges
them into one answer. Optional live web grounding and a debate round where models
cross-examine each other first.

Create App — describe an idea in plain English. ForkScape plans the tasks, designs the
screens, dispatches worker agents that write real files to a folder you choose, and starts
the app. Then change it by asking.

Design Studio — mock up every screen as real HTML before any code exists, refine with
plain words, and hand the design to the build agents.

Agents with hands — read, search, and edit files, run commands, call APIs. Every write
asks permission first and shows you the diff.

Memory Palace & Vault — facts and decisions persist across sessions and are mirrored to
plain markdown you can open in Obsidian, grep, or commit to git.

The honest meter — a dashboard that measures every real call: tokens actually sent,
cost per model, and savings versus a linear session doing the same work. A built-in
benchmark runs that comparison live on a topic you choose, so you can verify the claim
rather than trust it.

Your models, your machine — Ollama for fully local and offline, or bring keys for
Anthropic, OpenAI, Gemini, Grok, and OpenRouter. Per-node model routing, or let the auto
router pick. ForkScape knows each model's real context window and warns before a run
would overflow it.

First run

Pick a provider on the welcome screen. The fastest zero-cost path is
Ollama — install it, pull a model, and ForkScape detects it
automatically. Cloud providers just need an API key, stored locally on your machine and
never sent anywhere but that provider.

Installing an unsigned build

These builds are not code-signed yet, so both platforms will warn you on first launch:

• Windows — SmartScreen: More info → Run anyway
• macOS — right-click the app → Open → Open (double-clicking will refuse)
• Linux (AppImage) — chmod +x ForkScape_1.0.0_amd64.AppImage before running

Signing certificates are planned; until then this is expected, not a malware warning.

Free vs Pro

Free is the full local core, not a time-limited trial: unlimited local models, branching,
importing, research with 2 models, agents, Design Studio, and the Vault, across 5 canvases.
Pro lifts the canvas limit, allows deeper agent nesting and unlimited research models,
and adds scheduled automations and the Telegram gateway.

Known limitations

• Builds are unsigned (see above)
• File access, agents, app building, and the Vault are desktop-only — the browser
  playground is for exploring the canvas, not for building
• OpenAI and Grok can't be used from the browser playground (their APIs block direct
  browser calls); all providers work in the desktop app
• Collaboration and social publishing are Labs previews, off by default