product

Same Output

Pillar ④ · Auto preview, telemetry, agentic debug.

The problem with “it works on my machine”

When a Node produces a change, the result needs to be visible to people who don’t have the project checked out. Today this is the gap that kills “vibe coding” remote workflows: you can chat into a 5-second-ping latency without noticing, but you can’t preview the result without one.

Manifoldone closes that gap by hosting the output.

What gets hosted

Project typeWhat’s hostedWhere it shows up
Web appBuild artifact + preview URLDashboard, mobile, deep-link
Game (web target)Playable build, browser-streamedDashboard, mobile, deep-link
Game (native target)Lightweight cloud-streamed demoDashboard, deep-link
Library / CLILatest test run + artifact downloadDashboard

Vercel for everything

For web projects this looks essentially like Vercel: every push, every PR, gets a preview URL. The difference is that the build was triggered by an agent, in response to a prompt, on a Node — and the URL gets routed back into the same Thread that produced the change.

Telemetry feeds back

The build is observed. Performance data, errors, usage signals are collected. They feed back to the agent that produced the change as input for the next iteration. The loop:

intent → Thread → change → preview → telemetry → next intent

is closed. The agent is no longer flying blind once code leaves its hands.

Agentic debug — the printf agent

When something breaks, a specialized debug agent is dispatched. Its job is small and specific: form a hypothesis about where the bug is, instrument the code with logs, run the failing path, read the logs back, and brief the human on what it found.

This is deliberately old-school. We had printf debugging for a reason — it works. Letting an agent do it at scale is one of the highest-leverage applications of agentic workflow we’ve found.

v0.1 scope note

This pillar describes the full vision. v0.1 ships only the docking + activity + Recap path — auto-preview and telemetry come in Phase 3. The reason this page exists today is that the topology design has to anticipate this layer to avoid late surprises.