MW LIFECYCLE CONTRACT · v1.0 · SAGE APPROVED ✓

The Lifecycle Contract

The clean developer surface over the MW engine. This is what external developers meet. Not the engine. Not the backend functions. This.

v1 Decision — Option B

v1 exposes 3 calls: mwWake() → mwRun() → mwSleep()

mwCommit() is internal to Run/Sleep for v1. Will be exposed in v2 once the pattern is proven.

⚡

mwWake()

→

⚙️

mwRun()

→

🌙

mwSleep()

The Three Calls

Loads the last known safe snapshot for this actor+tenant pair. Reconstructs working state deterministically — no inference, no guessing. Returns a session token the caller uses for all subsequent calls in this session.

Accepts

tenant_idstringrequired

Derived from Bearer token server-side. Never from payload.

actor_idstringrequired

Which AI agent or actor is waking. e.g. 'arc', 'companion_id'

profilestring

Boot profile: 'builder' | 'companion' | 'agent'. Default: 'builder'

context_hintstring

Optional natural language hint to guide hydration. e.g. 'Focus on architecture decisions'

Returns

session_id

The only credential external developers need. Pass this to mwRun() and mwSleep().

snapshot_id

ID of the snapshot that was loaded. null if first session.

resumed_from_prior

boolean — true if a prior state was found and reconstructed. This is the continuity proof.

boot_packet

The reconstructed working state. JSON object. What the AI reads to orient itself.

latency_ms

How long wake took.

Maps to (internal engine)

mwBootstrap

Creates ContinuitySession record, issues wake token, generates verifier challenges

mwHydrate

Loads snapshot + working set memories, builds boot packet

mwCompose (optional)

If additional context injection needed — e.g. specialist routing

Failure behaviour

If snapshot is corrupt or reconstruction fails, MW halts and returns a recoverable error with the last safe snapshot ID. Never silently continues with bad state.

mwCommit() — v1 Internal Only

💾mwCommit() — v1 internal only

Mid-session state commit. Validates and persists a delta without closing the session. Allows long-running sessions to checkpoint safely.

v1 Decision

NOT exposed in v1 public surface. Internal to mwRun/mwSleep for now. Will be exposed in v2 once the pattern is proven in the wild.

Why

External devs on v1 don't need mid-session commits — Wake → Run → Sleep is sufficient proof of continuity. Adding mwCommit() to v1 introduces complexity without adding first-integration value. Ship Option B.

Internal Engine Map

The wrapper does not rebuild the engine. It wraps what exists.

Wrapper callEngine functionsWhat it does

mwWake()mwBootstrap + mwHydrateCreates session, loads snapshot, builds boot packet

mwRun()mwRunWithContinuityExecutes work through Guard-aware path, stages mutations

mwSleep()mwHandoverWrites delta, closes session, updates snapshot pointer

mwCommit() (internal)mwHandover (partial)Delta write without session close — v1 internal only

What This Contract Does NOT Do

✗

It does not rebuild the engine. The engine exists. This wraps it.

✗

It does not expose internal engine functions to external developers.

✗

It does not require new entities — ContinuitySession already exists.

✗

It does not expose mwCommit() in v1 — internal only until pattern is proven.

✗

It does not promise perfect symmetry with mwBootstrap/mwHandover naming.

For Sage — Pressure Test

This is the proposed wrapper contract. Before implementation begins, the questions we need answered:

Q1: Is the 3-call v1 surface sufficient to prove continuity?

✓

Yes. Wake → Run → Sleep is exactly right. mwCommit() internal for now.

Q2: Is the proposed_mutations model correct for v1?

✓

Conceptually yes, too explicit for v1 public surface. Expose intent (input), not mutation mechanics. REMOVED from public API.

Q3: Should wake_token be exposed externally?

✓

No. External devs only see session_id. wake_token stays internal. REMOVED from public returns.

Q4: Does anything require rebuilding the engine?

✓

No. Everything wraps existing engine functions. Do not modify engine unless wrapper makes it impossible.

Sage's verdict: Contract is correctly minimal. Design choices are good. Ready for Step 2 — implementation.

Contract locked. The golden path.

understand MW → run Hello World → see continuity proven → explore deeper APIs

🌍 Hello World Demo BetaKit Core Primer