For builders done babysitting their AI
Give your AI an
operating system.
AI coding is powerful — but it still makes you babysit the parts around the build. Claude Code and Codex build the app. LifeSpace gives them the operating layer they keep forgetting to maintain.
Don't forget.
Read the docs again.
Where's the backlog? Where are your notes?
Are you reusing what you already built?
Did the context window compact and lose the plan?
Where are my API keys — why did you paste one into chat?
LifeSpace exists so those questions have a default answer. Memory goes in Memory. Tasks go in Projects. Specs and decisions go in Knowledge. Credentials go in Keys — never your chat. Your AI redeems the whole layer from a single URL, and every fresh session picks up where the last one left off.
§ 01 · Every question, a home
The things worth keeping don't live in the chat scroll.
The babysitting happens because the work has nowhere durable to live. Give each thing a home and the nagging stops.
§ 02 · The operating set
Nine services that give the AI somewhere to put identity, credentials, memory, work, messages, and handoffs.
This is the working layer I use every day. Tenants and Trust tell the AI who it is, what tenant it belongs to, and which modules it can use. Keys gives credentials a safe home. Orchestrator coordinates the work. Projects, Memory, and Knowledge keep the plan, rules, and decisions alive. Dispatch sends messages. Handoff moves work between sessions without starting over. Together, they answer the questions AI coding keeps forcing the human to ask.
"Who is this AI working for?"
The single source of truth for the tenant tree — every service reads the same identity, so an AI is unambiguously scoped to one owner.
"What is it authed for?"
Shared auth and authorization — JWTs, SSO, and module-scoped grants. Every service delegates the "are you allowed?" question here.
"Where are my keys — why did you paste one into chat?"
A centralized, encrypted credential vault. The secret reaches the tool that needs it without ever appearing in a chat window or a repo.
"Who's doing what, across sessions?"
The durable control plane above the work — the implementation log, session registry, backlog, and activity, persisting state that any single session can't.
"Send that — but where did it actually go?"
One send endpoint for email and SMS, with sending routed correctly and every message tied back to the session that sent it.
"Pick up where the last session left off."
Structured work packets that move a task between sessions or agents carrying its spec and state — the convention for transferring work, not re-explaining it.
"Where's the backlog — what's built, what's still open?"
A real backlog of features and issues with native status and priority — the work lives in a queryable place, not buried in a chat transcript.
"Don't forget — and don't lose the plan when the context compacts."
Typed, scoped memories — rules, preflight checks, feedback, project facts — that the AI pulls back in automatically at the start of every session.
"Read the spec again."
An AI-first markdown store for specs, decisions, and conventions — the canonical knowledge base for a tenant, not docs scattered across Drive and Notion.
With the operating nine, nineteen modules are live in production today — plus MCP and Capture in use, and five more planned. Status reflects the live module catalog as of this build.
Guardrails are the product
The controls aren't a feature. They're the whole point.
Anything can call an API. The hard part — and the durable part — is the layer that decides who's allowed to, keeps the receipts, and stops things before they run away.
Tenant isolation on every row
One tenant can never see or touch another's data. Isolation is enforced at the data layer, not bolted on at the edge.
Encrypted credential vault
Provider keys live in an AES-256 envelope-encrypted vault with full access logging — never in a repo or a scattered .env file.
Full audit trail
Every credential access, message sent, and work handoff is logged to a tenant-scoped event bucket and is queryable.
Rate limits & spend caps
Budgets are enforced per tenant so an agent can't run away. The cost of autonomy stays bounded by design.
Human approval gates
Anything that spends money or goes public pauses for a person. The model proposes; a human releases the brake.
Scoped, expiring access
Onboarding grants a module-scoped token with a short life — access is narrow by default and dies on a timer.
§ 03 · Getting in
Three ways to dive in.
Let your AI read it
Paste one onboarding URL into Claude Code. It carries the MCP install steps and a scoped token — your AI wires up the tools and gets full access to them with zero extra setup from you, then can brief you on the platform. The AI-native way, and it proves the whole thesis in about ninety seconds.
Build in a sandbox
An isolated tenant that can't see or touch anyone else's data. Save a memory, track a task, send yourself a message — feel the guardrails in a few minutes.
Read the architecture
The North Star, the module specs, and an honest "where it still needs a human" brief. Open docs, an open MCP client; the rest is one short read away.
Preview access
Get a look inside.
Preview seats go out a few at a time into isolated sandboxes. Tell us what you want to build and we'll scope a tenant to match. Honest feedback on where it breaks is the whole point.
Got it — thank you.
You're on the preview list. We'll reach out with a sandbox URL and a short Start-Here. In the meantime, the architecture is yours to read.