EchoVault v0.5.0 — Living Memory for Coding Agents

EchoVault v0.5.0 turns persistent memory into a continuously maintained project working model for coding agents. Claude Code and Codex can now begin work with relevant decisions, verified procedures, known fixes, constraints, and unresolved tasks already in context—reducing repeated discovery and unnecessary tool round trips.

Highlights

• Added typed living memories for playbooks, known fixes, constraints, project state, and active work.
• Added task-aware, token-budgeted context packs that balance relevant decisions, unresolved work, active bugs, reusable patterns, and recency.
• Added structured playbooks with triggers, prerequisites, steps, verification, follow-ups, constraints, rejected alternatives, and open questions.
• Added provenance and validity metadata, including confidence, validity windows, commit, branch, links, verification timestamps, and supersession.
• Added an agent-level switch to enable or disable proactive context retrieval.
• Refreshed project-local integrations for both Claude Code and Codex.

Retrieval quality

• Added evaluation against redacted golden datasets with recall, ranking quality, irrelevant-result rate, latency, and context-token metrics.
• Added relevance calibration sweeps with separate lexical, vector, and hybrid thresholds.
• Added explicit empty results when no memory is relevant.
• Added raw scoring diagnostics through --explain.
• Added local retrieval feedback counters to support future ranking improvements.

Memory maintenance

• Added human-reviewable proposals for duplicates, contradictions, stale context, completed follow-ups, and overly broad session memories.
• Added memory doctor checks for vault/index drift, vectors, references, lifecycle issues, growth, latency, and configuration risks.

Install

pip install git+https://github.com/mraza007/echovault.git@v0.5.0
memory init
memory setup claude-code   # or: memory setup codex

See the README for the full living-memory workflow and configuration options.

v0.4.0 — Rust Dashboard Rewrite

What's New

The terminal dashboard has been completely rewritten in Rust (ratatui + crossterm), replacing the Python/Textual implementation.

Highlights

• 3MB standalone binary — instant startup, no Python runtime needed
• k9s-style keyboard navigation — 1-4 switch panels, j/k/g/G vim nav
• $EDITOR integration — press e to edit memories in vim as YAML
• Project filter — press p to filter by project across all panels
• : command palette and ? help overlay
• Async duplicate detection — UI stays responsive during O(n²) comparison
• Removed Textual dependency — smaller Python install footprint

Install

Python package (CLI + MCP server):

pip install git+https://github.com/mraza007/echovault.git@v0.4.0

Dashboard binary (macOS arm64 — download from this release):

curl -L -o ~/.local/bin/memory-dashboard \
  https://github.com/mraza007/echovault/releases/download/v0.4.0/memory-dashboard-darwin-arm64
chmod +x ~/.local/bin/memory-dashboard
codesign --force --sign - ~/.local/bin/memory-dashboard

Or build from source:

cd dashboard && cargo build --release
cp target/release/memory-dashboard ~/.local/bin/
codesign --force --sign - ~/.local/bin/memory-dashboard  # macOS only

Then run: memory dashboard

Full Changelog

See CHANGELOG.md

v0.2.0

Added

• Added memory dashboard, a Textual terminal dashboard for vault-wide browsing, editing, archive/restore flows, duplicate review, import, and reindex operations.
• Added archive-aware lifecycle metadata for memories, including archived state and merge provenance.
• Added stable markdown memory IDs so existing session files can be safely edited and rewritten.
• Added dashboard and lifecycle regression coverage for the new TUI and archive/merge flows.

Changed

• Reworked markdown session handling from append-only helpers into a round-trippable parser/writer that preserves session structure while supporting edits.
• Updated SQLite and search behavior so archived memories are excluded from normal search and listing paths by default.
• Improved memory import deduplication to key on (project, file_path, section_anchor) and hardened import parsing for legacy/BOM/CRLF markdown.
• Documented the new dashboard command in the README.

Fixed

• Fixed import behavior for same-title memories across session files.
• Fixed import decoding for legacy cp1251 and UTF-8 BOM/CRLF markdown inputs.