v0.0.3

v0.0.3

v0.0.2

0.0.2 is the initial release.

👀 Highlights

Observation tracking for research agents, backed by SQLite. Record raw signals, connect them, detect duplicates, promote what matures into hypotheses. Works as a library, a CLI, or through AI SDK tools.

👁️ Typed Observations

Five types (pattern, anomaly, measurement, correlation, artifact) with source tracking, confidence scores, frequency counting, and quality metrics. Full lifecycle: active, promoted, dismissed, archived.

🔗 Relations & Clusters

Seven relation types between observations (similar_to, contradicts, supports, derived_from, duplicate_of, refines, same_signal_as). Named clusters for grouping related signals.

🔍 Analysis & Search

FTS5 full-text search across titles, descriptions, and tags. Triage ranking by composite score. Queries for frequent, unpromoted, isolated, and convergent observations.

🧩 Dedup

Exact fingerprint, trigram similarity, and sourceRef matching. Scan, review, merge workflow.

⌨️ CLI

32 subcommands with --json and --toon output. Batch import/export for agent pipelines.

🤖 AI SDK Tools

Seven tools via obsxa/ai for AI SDK integration: observation, relation, cluster, dedup, search, analysis, promote.

📦 Data Safety

Drizzle migrations with auto-backup before schema changes. Schema version validation on startup. Manual backup/restore via CLI.

✅ Upgrading

pnpm add obsxa

👉 Changelog

compare changes

🏡 Chore

• tooling: Sync with template-ts (#2)