Unified Knowledge Layer
- Markdown files as canonical source of truth
- Typed frontmatter with immutable IDs for every object
- Workspace-scoped memory boundaries (personal, business, project)
- Obsidian-compatible — opens existing vaults without migration
Knowledge Operating System
Vault
The knowledge layer for the AI era.
Your AI agents are brilliant individually. Together, they're amnesiacs. Vault is the markdown-canonical, policy-governed memory substrate every agent reads, writes, and reasons against.
Every agent arrives with its own memory store — files in one folder, JSONL in another, a vector database somewhere else. The result is fragmented knowledge, inconsistent retrieval, and zero shared context — with no audit trail and no policy enforcement.
Not another vector database with a UI on top.
Not a cloud silo you rent until the vendor pivots.
Not free-for-all agent access with no governance.
One substrate. Every agent.
Vault replaces N fragmented memory stores with one durable, policy-governed, evidence-backed knowledge system. Files you own. Policies you write. Audit logs that can't be edited away.
What You Get
A governed memory substrate for every agent in your stack.
Four load-bearing capabilities, enforced in the file contract — not bolted on as middleware.
Policy-Governed Execution
- Non-bypassable execution gateway for all agent actions
- Approval tiers: read, propose, write, move, delete
- Least-privilege policies scoped per workspace, space and folder
- Append-only audit log for every agent interaction
Hierarchical Retrieval
- Workspace → Space → Folder → Structural → Semantic
- Structure and metadata are priors, not optional garnish
- Policy-aware at every boundary — retrieval respects access controls
- Semantic search as fallback, not the only retrieval path
Multi-Agent Integration
- Claude, Codex and specialized agents share one substrate
- MCP-native interfaces for seamless tool integration
- Typed agent and skill files with frontmatter contracts
- Each agent keeps its identity; Vault unifies the memory
5
Retrieval layers
100%
Local-first ownership
N
Agents, one substrate
0
Vendor lock-in
Architecture & Security
Built for trust, durability and safety.
Engineered so your data never leaves your control — and every AI action is governed, auditable and reversible.
Vault crate map
Tauri 2 Desktop App (TypeScript UI + Rust backend) │ ├── vault-core workspace, identity, file contract ├── vault-file-engine isolated file I/O with safety boundaries ├── vault-index structural + semantic indexing ├── vault-retrieval 5-layer hierarchical retrieval ├── vault-policy policy engine, ACL enforcement ├── vault-gateway non-bypassable execution gateway ├── vault-orchestrator agent coordination & routing ├── vault-providers LLM & embedding provider abstraction └── vault-sync operation-based sync & collaboration
Local-First
Every vault has a complete local replica. The cloud augments sync and sharing but never becomes the sole source of truth. Your files, your disk, your control.
Policy-Governed Gateway
Every agent action — read, propose, write, move, delete — flows through a non-bypassable gateway. Destructive actions require explicit approval. Policies are machine-readable YAML.
Evidence & Audit
Append-only audit logs record every agent interaction. AI-generated content carries provenance metadata, evidence references and freshness timestamps.
Tauri 2 + Rust
Desktop runtime
TypeScript
UI layer
Markdown + YAML
File contract
MCP Native
Agent protocol
First-Class Operations
Three operations anchor everything agents do.
Every agent interaction with your vault is an Ingest, a Query, or a Lint — or a compound of them. Simple vocabulary, powerful outcomes.
Ingest
Absorb a source, extract key insights, update relevant pages with evidence references, and append a curated log entry. The default entry point for new material entering your vault.
Query
Ask against the vault with hierarchical retrieval, synthesize an evidence-backed answer, and optionally file the result back as derived content. Every answer shows its sources.
Lint
Active health pass across a workspace. Detects contradictions, stale evidence, orphan pages, missing cross-references and broken links. Produces a report; never modifies content directly.
Getting Started
From install to governed knowledge in minutes.
Vault works on any folder. Point it at an empty directory or an existing Obsidian vault and you're running immediately.
01 Day 1
Install & Open
Download the Vault desktop app. Point it at any folder — empty or existing. Vault initializes workspace metadata without touching your files.
02 Day 2–3
Organize & Index
Promote folders to Workspaces or Spaces with a single frontmatter declaration. Vault builds structural and semantic indexes in the background.
03 Week 1
Connect Agents
Integrate Claude or other agents via MCP or directory access. Define policies per workspace. Every agent action flows through the governance gateway.
04 Ongoing
Compounding Knowledge
Every ingest, query and decision adds to a compounding knowledge artifact. Your vault grows smarter over time — with full audit trails and evidence at every step.
Your knowledge deserves a substrate, not a silo.
Vault is pre-alpha and shipping with select Cross Systems engagements. Tell us about your team and we'll send back an early-access scope.
Request Early Access