appendix

FAQ

Common questions, with the answers we believe today.

Do I need to install anything?

Yes — one process called Node, on the machine you want to dock. In the early beta this is npx manifoldone dock. Easier installation methods (single binary, drag-and-drop) are on the path to Phase 1 launch.

Is my code uploaded anywhere?

No. The Node runs on your machine. Source code stays on your LAN. The only things that travel through the relay are intent (what you typed) and results (what the agent produced as text). Files are touched locally by the agent, on your machine.

When a Thread ends, a one-paragraph Recap is generated through a Manifoldone-managed LLM gateway. The events sent to the gateway are Thread metadata (your prompts, tool names, file paths touched) — not raw file contents. Only the resulting summary text is stored in our database.

What about Perforce?

Supported in the long-term architecture. v0.1 ships Git only — Perforce follows when there’s a design partner with an actual game project to validate against. The same operations the system performs against Git (commit, branch, pull) translate to Perforce equivalents (changelist, branchspec, sync) by a per-VCS adapter. See Sync.

How is this different from Cursor?

Cursor is an AI-augmented IDE for an individual developer. Manifoldone is a coworking layer for teams whose individual developers already have Claude Code (or similar). The two are complementary in principle, though we expect most Manifoldone users to keep their IDE closed most of the time.

How is this different from GitHub?

GitHub is the system of record for code. Manifoldone is the system of record for the process that produced the code. Diffs go to GitHub; intent + chat + recap stay in Manifoldone.

What about latency?

A 5-second round-trip between your device and your Node is invisible to vibe coding because the agent’s thinking time absorbs it. We do not require dedicated low-latency network connections. We do require enough bandwidth to stream text — a few KB/s is enough.

Does it work offline?

The Node does. The local agent can keep working on your machine even if the relay is unreachable. Reconnection re-syncs state. The web/mobile client requires the relay to be reachable; we don’t yet have an offline UI.

What if my CC license is per-user?

Today we run one CC license per Node. In the single-user Phase 1 case, that’s your license, on your Node. The team Phase 2 case is structurally different and is the reason license virtualization is an explicit blocker on the roadmap.

Can I self-host?

The Node, yes — it’s already on your machine. The relay, eventually — Phase 3 introduces an on-premises relay image for compliance-sensitive customers. Memory’s project-local layer (Phase 2+) is a .memory/ directory committed to your VCS, so portions of it are already self-hosted by virtue of living in your repo.

What’s the relationship to estelle and soulstream?

estelle gave us the “Pylon = SSOT, Relay = stateless, Client = display-only” three rules and proved them for the single-user case. We renamed Pylon to Node and extended the topology to multi-Node — territory estelle’s README explicitly leaves untested.

soulstream gave us the HTTP/SSE event vocabulary that lets external bots integrate with a Claude Code conversation over the network. We use the same envelope on our internal wire so Soulstream-style integrations work against a Manifoldone Node almost for free.

Full credit in Inspirations.

When can I try it?

Sign up at the early access form on the landing page. We are letting in small teams in batches as Phase 1 stabilizes.