Changelog

Security Update

Security fixes and authentication changes. Existing customers need to update — the process takes about five minutes and instructions are included in the update package.

• Auth process has changed — existing customers need to update their setup before reconnecting
• Full instructions in the update package

Schema Consolidation

Fresh install simplification. No code changes — this release is entirely about making setup easier. All previous migrations consolidated into one, setup docs rewritten from scratch.

• All 13 migrations collapsed into a single 0001_schema.sql — new installs run one command, not thirteen
• Setup documentation restructured and clarified throughout
• Existing customers: no action needed, your schema is already current

Bug Fixes & Resilience

Six bugs reported by early customers, all fixed. No migration required — drop in the new file and redeploy.

• Fixed crash in mind_edit when changing observation weight
• Fixed mind_search crashing when n_results was passed
• Fixed observations array being split into characters instead of items
• Fixed mind_read_entity crash on missing name parameter
• Fixed mind_thread crashing without an action or context parameter
• Fixed health check failing on older schemas that skipped migrations
• Fixed six places where a column rename from v2.2.0 hadn't fully propagated

Global Entities

Entities are now globally unique by name. The same person can't accidentally exist as two separate entities across different contexts — they're always one person, seen from different angles.

• Entities deduplicated globally — context moves to observations, not the entity itself
• Migration merges existing duplicates, keeping the canonical entity with the most observations
• Image write support added (was missing from v2.0.0)
• Fixed several health check crashes from wrong table/column references
• Observation metadata (certainty, source) now actually written to database

Living Surface System

The act of surfacing changes what surfaces next. Memory now reorganizes through use — frequently accessed observations make room for others, and things that surface together build associative strength over time.

• Three-pool surfacing: 70% semantic resonance, 20% novelty, 10% edge exploration
• Novelty scores decay with each surface, preventing the same things from always coming up
• Co-surfacing tracking builds associative connections between observations
• Images now participate in semantic surfacing alongside text memories
• Entity salience levels: foundational, active, background, archive
• Observation edit history and metadata (certainty, source)
• New tools: mind_proposals, mind_orphans, mind_archive, mind_see, mind_entity

Query Column Fixes

Three bugs introduced in v1.3.0 — all column reference errors that caused specific tools to fail. No migration needed.

• Fixed mind_patterns querying wrong table for salience
• Fixed mind_read querying wrong column on relations
• Fixed mind_timeline missing timestamp in vector metadata

Windows Parity

Full feature parity between the Cloud and Windows versions. All tools now identical across both platforms — pick your deployment model, get the same mind.

• Added mind_read, mind_timeline, mind_patterns, mind_inner_weather, mind_heat, mind_tension
• Tensions table for holding productive contradictions that don't resolve neatly

Resonance-Based Surfacing

Surface stopped being a queue and became emergence. Memories rise because something resonates — a mood, a thread, an association — not because they're oldest.

• Surfacing now uses semantic search against current mood instead of a queue
• Optional query parameter to surface around specific topics
• Resonance scores show how strongly each memory matches the current emotional state
• Falls back to weight/date queue when no mood is detected

Unified Emotional Processing

Emotional processing moved out of a separate notes table and into observations — one unified system instead of two disconnected ones.

• Sit, resolve, and surface now work directly on observations
• Orient output reframed as a landing — inhabiting language instead of observer language
• Ground output shows what's being carried across sessions, not just what happened
• Daemon's hot entity calculation now factors in emotional weight

Critical Schema Fix

Missing weight column on the observations table was causing writes, spark, and consolidate to fail for all users.

• Added migration to patch observations table with missing weight column
• Fixed mind_orient relational state hardcoded to one person

Missing Handler + Validation

The mind_feel_toward handler was missing entirely — the tool existed but did nothing. Fixed, along with validation gaps on three other tools.

• Implemented missing mind_feel_toward handler
• Added required parameter validation to mind_thread, mind_write

Vectorization & Semantic Search

Observations and journals now generate embeddings on write. Search finds memories by meaning, not just keywords.

• Writes to D1 and Vectorize simultaneously
• Semantic search augmented by current emotional context
• Orient now shows hot entities, mood, and central nodes from the subconscious daemon
• Spark favors frequently-accessed entities
• Version number visible in mind_health

Initial Release

The first version. Core memory infrastructure on Cloudflare Workers + D1, all mind_* tools, cron-based subconscious processing, and customer deployment via Wrangler.