FAQ

Why not just use my AI's built-in memory?

Built-in memory is per-client and per-account: it doesn't move a project's full working state across a new thread, a different model, or a different provider without drift. Soil's unit is a graded, drift-checked handover you can save in one client and load in another · and it stores only safe projections, never raw chat.

Does Soil read my chat history?

No. A Remote MCP server can't read your conversation · only the AI already in the room can. Soil directs that model to extract against its recipe, then checks and grades the result.

What does Soil store?

Only a safe projection · the meaning a fresh model needs, never the secret. Raw prompts, provider payloads, secrets, source contents, and private paths are never stored; a fail-closed guard enforces it.

Can I control how much detail a handover captures?

No, and that's deliberate. The recipe is comprehensive by design, and Soil adapts the depth to the project and checks a cold model could continue. There's no depth dial · the engine decides.

What are the working-style questions at save time?

Every save asks the saving model a few situational working-style questions · for example, "an unexpected blocker appears in this project tomorrow: what does this project do first, and why?" The answers are recorded observations: stored raw with the save, never scored or graded, never blocking a save, and shown as a clearly labeled block when the handover is loaded. Two questions are fixed and two rotate across a project's successive saves, so a project covers the full set over time. The mechanism is free on every save; a whole-project load also shows the observations of each thread it restores in full, attributed per thread.

How do I sign in?

With Google or with email and password. Email sign-ups confirm with a verification email, and there's a password reset if you forget it. The same identity works across the console dashboard and the MCP connection, so the same login across different clients shares the same handovers. Connections renew silently once made, and you can revoke any client from the console's Connect page.

What's the difference between a personal save and a project?

A personal save (soil save, or soil save #my-name) is a standalone snapshot · complete on its own, free. Use it for "save this state now, load it later or elsewhere." A project is where your work accumulates over time: each soil save @project adds a thread, and soil load @project brings the project back · the project knowledge (its decisions, constraints, and architecture) in full, every thread listed and attributed, the newest restored inline, and any older one readable on demand with soil load @project #NNN. The more you save into it, the more complete the project knowledge becomes · a compounding, restorable brain a fresh AI or a teammate can pick up. Names are always kept, never overwritten, so nothing you saved is ever lost. Owning a project is the paid capability; see projects & saving over time.

Can I share handovers with a team?

Yes, with projects. Create a shared project, invite others (by link or by email), and everyone can save their own threads into it (soil save @project #auth-flow to name the thread, or soil save @project to auto-name it), see what's there (soil list @project), and restore a single thread (soil load @project #auth-flow, by name or by code) or the project in one restore (soil load @project). Personal saves and loads stay free; owning a project is a Pro feature.

How does Soil prove a handover actually works?

By a round-trip on your own two models: the model you save from writes the handover plus the answer key (it's the only thing that still has the full context), and the model you load into answers cold; Soil scores the answers deterministically · pass, unknown, or wrong. Quality is measured, not claimed. The save-time grade is a forecast; the measurement happens at load, and today it runs on project loads (soil load @project). A personal (#NNN) load carries the save-time readiness grade only · load-time measurement there is on the roadmap.

What if Soil's tools disappear in ChatGPT?

In a very long or old ChatGPT chat, ChatGPT can stop exposing the connector's tools to that chat · the model says it can't run soil save. It's a ChatGPT limitation, not a Soil outage. Usually you just start a new chat and save there. If the thread still holds context you can't recreate, you can rescue it: Rescue a dead ChatGPT thread turns the stuck thread into a normal handover with its own load code.

How do I know a save actually succeeded?

By the receipt. A save (or import) counts only when you get back a numeric #NNN load code · that card is the proof. No #NNN means it didn't save, no matter what the chat says. Your console dashboard and soil list are the source of truth: if it isn't there, it didn't save.

What if my client can't connect over MCP?

There's always a path. Open the console dashboard, copy a handover's restore prompt, and paste it into any tool · the copy-paste backstop works everywhere, including non-MCP clients.

Is the Desktop app available?

Yes · Soil Desktop is live in beta, alongside Soil Cloud (the Remote MCP connector). Desktop is the local, offline-first app: your handovers live on your machine, an AI client connects to a local MCP server, and cloud backup is opt-in.