00CHANGELOG

Every shipped change, in order.

WorldMind is in technical preview. This log records every change that has reached a partner, even if it is not public. Dates are UTC. Everything below v0.1 was shipped to internal testing only.

v0.1.0
2026-04-20
SHIPPED

Technical preview announced

  • Public landing and manifesto published.
  • Private partner program opens. First four studios begin integration.
  • Runtime core exposes POST /v1/agents, /v1/events, /v1/decisions.
  • Five memory layers online: episodic, semantic, relational, procedural, scratchpad.
  • Structured output contract enforced end to end. Schema violations trigger repair.
v0.1.0 · SHIPPED2026-04-20
v0.0.9
2026-03-28
INTERNAL

First vertical demo in a partner MMO

  • Quest NPC demo running against a private branch of a partner MMO.
  • Two memory profiles shipped: npc_social_basic, npc_faction_aware.
  • Trace dashboard v1. Shows context size, memory hits, latency, repair count.
  • Fallback policy is now mandatory at agent registration.
v0.0.9 · INTERNAL2026-03-28
v0.0.7
2026-03-03
INTERNAL

Tactical planner and vehicle doctrine

  • Tactical brain ships behind feature flag for BeamNG bridge.
  • Route abstraction over nav mesh. Agents reason about segments, not meters.
  • Per-second planning loop with utility AI fallback on timeout.
  • Capability allow-lists are now enforced server-side.
v0.0.7 · INTERNAL2026-03-03
v0.0.5
2026-02-11
INTERNAL

Validator + repair loop

  • Action validator landed. Anything outside the capability registry is rejected.
  • Repair loop re-asks the model with a diff of what failed. Retries capped at two.
  • Structured output error rate dropped from 8% to 0.4% on the internal test harness.
v0.0.5 · INTERNAL2026-02-11
v0.0.3
2026-01-22
INTERNAL

Memory write pipeline

  • Event salience scoring: ignore, episodic, or semantic promotion.
  • Relational memory with trust, grudge, fear, debt, attraction deltas.
  • First doctrine pack ('merchant_v0') ships as a testbed.
v0.0.3 · INTERNAL2026-01-22
v0.0.1
2026-01-04
INTERNAL

Repo exists

  • Context builder runs against a synthetic game state.
  • Model router chooses between tiny, mid, and strong tiers.
  • No memory yet. No validator yet. Just a loop.
v0.0.1 · INTERNAL2026-01-04
NEXT ON THE BENCH

Planned

  • PLANNEDAuthoring studio. Persona editor, doctrine packs, memory policy UI.
  • PLANNEDShared faction memory. Event propagation across agents in the same settlement.
  • PLANNEDBatched cognition for RimWorld-like populations.
  • PLANNEDMulti-region hosted inference.

RSS · atom feeds coming in v0.2