Vault Agent Memory v0.7.30

Vault Agent Memory v0.7.30

v0.7.30 is the post-rename packaging and branding patch for Vault Agent Memory.
It does not change the memory-governance model; it makes the published package,
active docs, OpenClaw integration, and repository links line up with the new
product name.

Highlights

• The canonical GitHub repository is now zycaskevin/Vault-Agent-Memory.
• Active README, guide, landing, OpenClaw, script, benchmark, and generated
  user-facing surfaces now use the Vault Agent Memory display name.
• Package metadata includes the new homepage, repository, and issues URLs.
• vault and vault-mcp help text now describe Vault Agent Memory.
• OpenClaw now shows Vault Agent Memory as the skill/plugin display name while
  preserving the compatibility id/path vault-for-llm.

Compatibility

The PyPI package name remains unchanged:

pip install --upgrade "vault-for-llm[mcp,supabase]==0.7.30"

The console scripts remain:

vault --version
vault-mcp --help

Validation

This patch was checked with the full local test suite, README command smoke,
OpenClaw adapter smoke, build/twine/wheel smoke, release parity, public PR gate,
and Release Readiness CI on the rename follow-up PRs.

Vault-for-LLM v0.7.29

Vault-for-LLM v0.7.29

v0.7.29 is an onboarding, safety, and demo-depth patch release for the
agent-assisted builder audience.

Highlights

• vault quickstart gives new users a small first-run path instead of starting
  with the full setup-agent flag surface.
• vault guide install now includes a 5-minute quickstart and FAQ for
  agent-assisted setup.
• Gateway startup, security headers, and graceful shutdown behavior are safer
  and more explicit.
• vault import obsidian --dry-run now has a tested no-write contract for
  active database rows.
• Memory export now supports Markdown, OKF, and JSON bundles with dry-run
  previews.
• Remote candidate sync can rotate HMAC keys with VAULT_SYNC_HMAC_SECRETS,
  and vault remote hmac-keys reports active key status without exposing
  secrets.
• The generated agent-governance demo now includes follow-up guides for
  consumer mode, automation mode, and multi-host remote candidate sync.

Install

pip install --upgrade "vault-for-llm[mcp,supabase]==0.7.29"

First Checks

vault quickstart --non-interactive --json
vault demo agent-governance --json
vault remote hmac-keys --json

Validation

This release line was checked with the full local test suite, README command
smoke, module-size gates, a clean Python 3.11.14 local package install smoke,
and Release Readiness CI on the onboarding/security/P3 PRs.

Vault-for-LLM v0.7.28

Vault-for-LLM v0.7.28

This is the public-preview readiness release for Agent-assisted builders.

Highlights

• Rendered three-agent memory-governance demo page, with English, Traditional Chinese, and Simplified Chinese switching.
• GitHub Pages deployment for the landing/demo site: https://zycaskevin.github.io/Vault-for-LLM/
• External contributor onboarding: issue templates, PR template, Code of Conduct, and good-first-issue ideas.
• Public positioning tightened around Agent-assisted builders and memory governance rather than a generic RAG, notes, or vector-search tool.

Changed

• Updated install examples to vault-for-llm==0.7.28.
• Updated package metadata to describe Vault as local-first memory governance for AI agents.
• Kept the honest boundary: Gateway/Supabase are adapters; this is not magic offline multi-master sync.

Validation

• Release parity check passed for v0.7.28.
• README documented command smoke passed.
• Targeted landing/setup/guide tests passed.
• Built wheel and sdist; twine check passed.
• Release readiness CI passed on PR #315.

Vault-for-LLM v0.7.27

Vault-for-LLM v0.7.27

This release focuses on the human-facing review surface for agent memory operations.

Added

• A gentler GUI review surface for Obsidian conflicts so Daily Report and Memory Control Center show the same human-review items.
• Gateway / Remote access health in the multi-Agent dashboard, including blocked/failed events, unique client IP counts, and next safe actions.
• Human-first Obsidian conflict inbox wording with three explicit choices: accept Obsidian, accept Vault, or keep both.

Changed

• vault daily-report --json now includes the standard ok: true success marker while preserving the existing report payload.
• GUI documentation now explains Dashboard review cards, Obsidian conflict inbox behavior, and Gateway / Remote safety expectations.

Safety

• Obsidian conflict inbox notes remain metadata-only and do not include conflicting note bodies.
• Gateway / Remote health remains read-only in the GUI and keeps blocked events visible before wider exposure.

Validation

• Full local test suite: 2378 passed, 10 skipped.
• Release PR CI passed on #296.
• Built wheel and sdist; twine check passed.
• Local wheel smoke verified vault --version, init, add, search --json, daily-report --json, and gateway audit --json.

Vault-for-LLM v0.7.26

Vault-for-LLM v0.7.26

Small release focused on the agent-facing CLI JSON contract.

Changed

• Normalized vault search --json with a stable ok / status envelope while preserving existing query, mode, count, and results fields for compatibility.

Validation

• Full local test suite: 2358 passed, 10 skipped.
• Release Readiness CI passed on PR #286.
• Built wheel and sdist; twine check passed.
• Local wheel smoke verified version, vault search --json envelope, and vault-mcp entrypoint.

Vault-for-LLM v0.7.25

Highlights

• Adds stable JSON output for vault init --json and pretty JSON output for vault init --pretty.
• Adds stable JSON output for vault compile --json and pretty JSON output for vault compile --pretty.
• Keeps compiler progress messages inside the JSON payload, so agent stdout stays parseable and automation no longer needs special handling for init/compile.

Validation

• PR #284 Release Readiness CI passed, including Python 3.10/3.11/3.12 tests, build/wheel smoke, README smoke, module size gate, privacy scans, Search QA regression gate, and uv workflow smoke.
• Local full test suite passed: 2357 passed, 10 skipped.
• Local targeted CLI/release tests passed: 59 passed.
• uv lock --check passed after syncing the version bump.
• Built v0.7.25 wheel and sdist locally; twine checks passed.
• Clean Python 3.12 wheel install verified package metadata and vault init --json / vault compile --json parseable output.

Notes

• Optional semantic dependencies remain optional; doctor may warn when ONNX/transformers are not installed.
• macOS system Python 3.9 is not supported; use Python >=3.10.

Vault-for-LLM v0.7.24

Highlights

• Adds safer Obsidian mirror conflict detection: when both the source note and Vault raw copy changed, Vault preserves both sides and records a metadata-only conflict for review.
• Adds generated Vault Remote Server hardening materials, including env-file templates and a deployment checklist for token handling, TLS, backups, audit review, and candidate-first remote writes.
• Adds a compact multi-agent Activity Health surface for connected agents, pending candidates, sync conflicts, Obsidian mirror conflicts, and daily human-review pressure.
• Bumps the release version from v0.7.23 because v0.7.23 was already published before these pre-release hardening changes landed.

Validation

• Release Readiness CI passed on PR #282 and PR #283, including Python 3.10/3.11/3.12 tests, build/wheel smoke, README smoke, module size gate, privacy scans, Search QA regression gate, and uv workflow smoke.
• Local full test suite passed before the version bump: 2356 passed, 10 skipped.
• Local targeted release tests passed after the version bump: release parity, CLI JSON contracts, setup-agent, gateway, Obsidian import, and GUI tests.
• Clean wheel install verified v0.7.24 metadata.
• Setup-agent generated Remote Server, Supabase realtime, Obsidian, n8n/Coze, and validation templates successfully.
• Generated local smoke test passed.
• Vault Remote Server was started locally and validated through the generated client: health, OpenAPI, search, and candidate-first memory submission passed.
• Twine package checks passed for wheel and sdist.

Notes

• Optional semantic dependencies remain optional; doctor may warn when ONNX/transformers are not installed.
• vault init and vault compile are functional but not yet part of the stable --json command contract; that remains a follow-up for stricter agent-first automation.

Vault-for-LLM v0.7.23

Vault-for-LLM v0.7.23

v0.7.23 makes Obsidian a better human-facing surface for Vault memory.

Highlights

• Obsidian can now act as the human review inbox for Vault.
• setup-agent can generate conservative Obsidian folder rules, import notes,
  schedule ongoing sync, and export review cards back to
  00-Vault-Knowledge/_Inbox/.
• The generated inbox includes:
  • Daily Memory Report.md
  • Memory Candidates.md
  • Sync Status.md
  • Folder Rules Preview.md
• Obsidian [[wikilinks]] are preserved during import and can become
  obsidian_link graph edges after vault graph build.

Why It Matters

Vault remains the governed memory core for agents. Obsidian becomes the place
where a normal user can read, skim, and review memory without learning the full
CLI.

Human notes in Obsidian <-> Vault governed memory <-> Agents via CLI/MCP/Gateway

Install

pip install --upgrade "vault-for-llm[mcp,supabase]==0.7.23"

First Checks

vault setup-agent --help
vault import obsidian --help
vault export obsidian --help

Validation

This release line was checked with Obsidian import/export tests, setup-agent
tests, module-size and history-privacy gates, build checks, and a real
setup-agent Obsidian sync smoke that wrote the generated _Inbox notes.

Vault-for-LLM v0.7.22

Vault-for-LLM v0.7.22

v0.7.22 focuses on making Vault easier to connect to many agents without
spreading memory across many tools.

Highlights

• vault gateway serve adds one small HTTP entrypoint for agents, scripts, and
  hosted-tool bridges that do not need the full CLI or MCP surface.
• Gateway supports health checks, search, bounded source reading, and
  candidate-first memory submission.
• Gateway is conservative by default: token authentication is on, agents must
  identify themselves, private memory stays hidden, and writes go to review
  candidates rather than active knowledge.
• vault import memory can preview and import memory from other tools or chat
  exports as candidates, including Markdown, JSON, CSV, transcript, and
  OKF-like inputs.

Why It Matters

Vault can now act more like one governed memory door:

Agent / workflow / device -> Vault Gateway -> reviewed local memory

The storage and transport adapters can evolve later, but users keep one
inspectable memory vault.

Install

pip install --upgrade "vault-for-llm[mcp,supabase]==0.7.22"

First Checks

vault gateway --help
vault import memory --help
vault doctor

Validation

This release line was checked with Gateway tests, memory migration tests,
project-dir tests, access-policy tests, MCP security tests, and the full local
suite: 2338 passed, 10 skipped.

Vault-for-LLM v0.7.21

Vault-for-LLM v0.7.21

v0.7.21 is a GUI and release-packaging polish release for normal Agent users.

Highlights

• The GUI now has one unified daily review inbox for the small set of items
  humans should actually inspect: daily report cards, candidate memories, sync
  conflicts, and directed Task Ledger handoffs.
• Directed Task Ledger handoffs can now be claimed from the GUI with explicit
  confirmation. This updates handoff state only; it does not write to L0-L3
  memory.
• Sync conflicts remain explicit: keep local, accept remote, or mark manual.
  Accepting remote promotes the reviewed candidate and archives the old local
  row instead of silently overwriting it.
• The GUI HTML is now packaged as vault/assets/gui_app.html, making the
  Python GUI module small and easier to maintain while keeping wheel installs
  working.

Safety Boundary

The unified review inbox is metadata-only by default. Raw candidate content,
sync-conflict content, and handoff Markdown stay hidden until the user opens
the specific detail view.

Install

pip install --upgrade "vault-for-llm[mcp,supabase]==0.7.21"

Validation

This release line was checked with GUI, Task Ledger, and multi-host sync tests,
module-size gates, wheel build checks, and a packaged GUI asset smoke test.