Changelog
What's New
The latest updates, improvements, and new features in CodeCortex. For detailed release notes, see GitHub Releases.
View releases on GitHubv0.5.0
changedClarified positioning: persistent knowledge layer for cold-start elimination. Not a middleware, not a proxy
changedTool count 15 to 13: removed analyze_module + save_module_analysis, renamed report_feedback to record_observation
changedAGENT.md rewritten with clear "when to use grep instead" guidance
changedCLI `modules <name>` output capped at 30 edges (was dumping 559KB for large modules)
addedValidated test data: ~50% fewer tokens, 2.5x fewer tool calls, same answer quality (23/25) on a 6,400-file codebase
removedDropped unvalidated "85% token reduction" claim. Replaced with real test numbers
v0.4.4
addedFreshness flags on all MCP read tool responses so agents know when knowledge is stale
addedget_edit_briefing tool for pre-edit risk assessment with co-change warnings, hidden deps, bug history
fixedHidden dependency warnings now only fire for files without import relationships (was falsely flagging imported pairs)
fixedCommand injection vulnerability in project discovery (replaced shell exec with direct file reads)
changedExtracted shared module-building logic to reduce code duplication
v0.4.3
fixedAdded Node 20-22 engine restriction. Node 24 not yet supported (tree-sitter upstream issue)
fixedFixed broken CodeScene whitepaper link in README
changedInstall command now includes --legacy-peer-deps in docs
v0.4.2
removedRemoved tree-sitter-liquid (git dependency caused native build failures on Node 22+). 27 languages now supported.
fixedGlobal npm installs now work reliably on Node 20-22
v0.4.1
addedGrouped CLI help: commands organized into Core, Query, and Utility sections for easier scanning
changedShorter, more scannable CLI command descriptions
changedUpdated npm package description to "Permanent codebase memory for AI agents"
v0.4.0
addedGit hook integration: `codecortex hook install/uninstall/status`. Auto-updates knowledge after every commit and merge
addedUpgrade command: `codecortex upgrade`. Detects your package manager (npm/yarn/pnpm/bun) and runs the correct install
addedUpdate notification: passive check on every CLI run shows a styled notification when a newer version is available on npm
v0.3.2
addedLiquid (Shopify) template language support (removed in 0.4.2 due to native build issues)
fixedForked and fixed tree-sitter-liquid upstream bugs: missing external scanner in binding.gyp and broken install script
fixedAligned version and language count across MCP server, JSON-LD metadata, site components, and README
v0.2.0
added`codecortex symbols [query]` to browse and filter the symbol index with kind, file, and exported filters
added`codecortex search <query>` for full-text search across all CodeCortex knowledge files
added`codecortex modules [name]` to list all modules or deep-dive into a specific module with dependencies
added`codecortex hotspots` to show files ranked by risk score (churn + coupling + bug history) with hidden dependencies
addedAuto-generated structural module docs during `init` and `update`. Includes exported API, dependencies, and temporal signals without requiring LLM analysis
fixedmodules/ directory no longer empty after init. Structural docs now generated for all detected modules
v0.1.1
fixedGo type extraction now traverses type_spec nodes inside type_declaration blocks, resolving missed type symbols
changedPublished to npm as codecortex-ai
changedUpdated all documentation to reflect native N-API binding. Removed outdated WASM references
changedCleaned up README: removed ASCII architecture diagram, expanded language support table to all 27 languages
v0.1.0
addedInitial release. Pre-digests codebases into structured knowledge served to AI coding agents via MCP
addedNative tree-sitter N-API parsing for 27 languages (TypeScript, JavaScript, Python, Go, Rust, C, C++, Java, Kotlin, Swift, and 17 more)
added14 MCP tools: 9 read (symbol lookup, dependency graph, git history, pattern search, decision log, session context) + 5 write (capture decisions, log sessions, tag patterns)
addedSix knowledge layers: Structural, Semantic, Temporal, Decisions, Patterns, Sessions
addedThree-tier memory model (hot/warm/cold) with ~85% token reduction for AI agent context windows
addedStreaming JSON writers for symbols.json and graph.json. Handles repos with millions of entries (tested: Linux kernel 5.3M symbols)
addedGit temporal analysis: commit hotspots, co-change coupling, file churn metrics
addedCLI with init, serve, and status commands
added136 vitest tests across 12 test files