v0.8.0

Full Changelog: v0.7.0...v0.8.0

v0.7.0

Full Changelog: v0.6.0...v0.7.0

v0.6.0

Full Changelog: v0.5.0...v0.6.0

v0.5.0

What's New

Highlights

• Fully local — removed all OpenAI/Ollama API dependencies. ctxgraph now runs 100% locally with no external calls.
• Relation classifier — new embedding + logistic regression classifier for relation extraction, with training pipeline.

Changes

• feat: add relation classifier (embedding + logreg) with training pipeline
• refactor: remove OpenAI/Ollama dependencies, position as fully local
• feat: intersection-verified relation extraction
• perf: relax API merge constraints, improve heuristic direction logic
• feat: improve extraction quality
• fix: update multitask model to gliner_multi-v2.1 ONNX
• fix: sentence-aware relations, MCP E2E tests, embed integration tests
• perf: embedding cache + traverse_batch tool to cut MCP hot-path latency

Full Changelog: v0.4.0...v0.5.0

v0.4.0 — Tier 2 + Git Watch (Launch Milestone)

Coreference resolution, fuzzy entity dedup (Jaro-Winkler), and ctxgraph watch --git for passive commit capture. 96 tests passing. Install: cargo install ctxgraph-cli

v0.3.0 — MCP Server + Semantic Search

MCP server + semantic search (all-MiniLM-L6-v2) + RRF fusion. Install: cargo install ctxgraph-mcp

v0.2.0 — Extraction Pipeline

What's new

• NER engine — GLiNER large v2.1 (ONNX, local, zero API cost) extracts entities from free text
• Relation extraction — heuristic pipeline + schema-driven co-occurrence scoring
• ExtractionSchema — TOML-driven entity/relation type configuration via ctxgraph.toml
• ExtractionPipeline — orchestrates NER → relations → graph insertion
• Model manager — download and manage ONNX models locally
• Benchmark corpus — 50 software-engineering episodes, 75 tests, F1 evaluation harness

Benchmark (GLiNER large v2.1, zero-shot, threshold 0.2)

Entity F1: 0.345 | Relation F1: 0.033 | Combined F1: 0.189

Target for v0.3: 0.800 via fine-tuned model or opt-in LLM extraction tier.

Install

cargo install ctxgraph-cli

v0.1.0 — Core Engine

Initial release of ctxgraph — local-first context graph engine for AI agents and human teams.

What's in v0.1.0

• SQLite storage with WAL mode and FTS5 full-text search
• Bi-temporal edges — valid_time + transaction_time for temporal reasoning
• Graph traversal — multi-hop walks with depth limiting via recursive CTEs
• Episode/Entity/Edge type system with fluent builder API
• CLI commands: init, log, query, entities, decisions, stats
• 24 tests covering all core functionality

Install

cargo install ctxgraph-cli