v0.3.0

Changes

• resume now includes a middle_context section sourced from context_bundle logic over the same conversation's filtered middle slice
• preserved first real exchange, explicit signal lines, and latest exchange in resume output
• kept extraction-noise filtering for AGENTS/environment boilerplate, skill-file reads, and write_stdin

Validation

• cargo test passed
• cargo build --release -p cli-memory-app --bin cmb passed
• npm pack --ignore-scripts produced aminovsky-cli-memory-0.3.0.tgz

v0.2.6

Changes

• filtered extraction noise from resume, context bundle, init, and refresh for Codex imports
• suppressed AGENTS/environment boilerplate, skill-file reads, and write_stdin from extracted conversation context
• kept first real user/assistant exchange and latest real exchange in resume bundles
• refreshed the local database with the patched binary before release validation

Validation

• cargo test passed
• cargo build --release -p cli-memory-app --bin cmb passed
• npm pack --ignore-scripts produced aminovsky-cli-memory-0.2.6.tgz

v0.2.5

Changes

• signal-only content extraction for imported CLI transcripts
• resume/context capture focused on opening exchange, latest exchange, tools, commands, skills, plans, and tasks
• context-bundle and conv-search fixes for exact conversation lookup and cold-start performance
• MCP startup auto-refresh and connection stability fixes
• provider importer noise filtering tightened across Claude, Codex, Copilot, OpenCode, Hermes, Zed, Gemini, and Antigravity CLI

Validation

• cargo test: 98 passed
• cargo build --release --bin cmb passed
• npm pack --ignore-scripts produced aminovsky-cli-memory-0.2.5.tgz

v0.4.0

Summary

• Prevent stdio MCP servers from killing peer sessions.
• Keep singleton/process coordination confined to future shared-mode startup only.
• Bump the release to 0.4.0 and add regression coverage for stdio coexistence.

Test Plan

• cargo test -p cli-memory-app --test cli_tests serve_stdio_instances_can_coexist_without_killing_each_other -- --nocapture
• cargo test -p cli-memory-app
• cargo build -p cli-memory-app
• npm pack --json

v0.2.4

Changes

• Fix 100% CPU at idle: Batch vector insertion computes turbovec codebook once instead of per-vector (9455× → 1×)
• Remove auto-exit: Server stays alive for MCP clients (OpenCode, Claude Code, etc.)
• Pre-load embedding model: Model loaded at startup for instant first-query response
• Version bump: 0.2.1 → 0.2.4

Resource Usage

Metric	Before	After
CPU at idle	99-100%	0.0%
Codebook computations	9455×	1×
Server stability	Exits after 1st tool call	Stays alive

v0.2.1 – Auto-exit server after tool call

Changes

• Auto-exit after tool call: The MCP server now exits after each tool call completes, instead of staying alive consuming resources. Uses CancellationToken + serve_with_ct for clean shutdown.
• The server is restarted fresh by the CLI on each subsequent call — refresh is fast enough that this is cheaper than keeping a long-running process.

Files

• cmb-0.2.1-darwin-arm64 — macOS ARM64 binary

v0.2.0

Breaking Changes

• Binary renamed from cli-memory to cmb — the Rust build produces a single cmb binary. The npm package still exposes both cli-memory and cmb aliases via the JS wrapper, so existing npx/npm invocations are unaffected.

Changes

• Add singleton process management — atomic PID file prevents duplicate server instances that drain CPU
• Add JSONC support — config files with // comments (Zed settings.json) now parse correctly
• init now warns instead of panicking on config install failures
• Update README with cmb command references

Fixes

• Fix panic when cli-memory init encounters JSONC config files (e.g. Zed)
• Fix duplicate cli-memory serve process accumulation across CLI sessions

v0.1.16

Changes

• Add cmb bin alias for shorter CLI invocation
• Expand install/unlink provider bundles (opencode, hermes, antigravity)
• Update cli-memory-retrieval skill with retrieval workflow rules
• Add bin_test.rs test suite for CLI validation
• Version bump to 0.1.16

v0.1.15

Release 0.1.15: rmcp MCP server, resume exact-source fallback, retrieval skill, and package metadata cleanup.

v0.1.14

feat: fully re-implement Python's hybrid search architecture (RRF). Merges dynamic keyword matching and HNSW semantic vector hits seamlessly with a 1.0/(rank+1) RRF scoring algorithm, while preserving background warmup protection.