Skip to content

One Memory for All Your Agents

Your context is trapped in silos. Obsidian here. Bookmarks there. Notes scattered everywhere. Every AI agent starts from scratch, asking the same questions, losing the same answers. Hoard changes that. One local index. Every agent. Instant recall.
πŸ“ note
πŸ”– bookmark
πŸ“„ file
πŸ”— link
πŸ’‘ idea
nom!
100%
Local
<300ms
Search
∞
Agents
1
Index

How It Works

Hoard syncs your local data into a single SQLite database with hybrid search, then exposes it through MCP. Connect once, access everything. Orchestrator (Beta) coordinates agents and workflows on top of the same local database.

Local-First

Your data never leaves your machine. No cloud. No tracking.

Universal MCP

Claude Code, Codex, any MCP client. Same data, same context.

Orchestrator (Beta)

Coordinate multiple agents, tasks, and workflows in one place.

Chunk Citations

Every answer traced back to its source. Stable IDs you can trust.

Hybrid Search

BM25 keywords + vector semantics. Fast and relevant.

Portable Memory

Switch agents, keep context. Your preferences follow you.

Instant

Sub-300ms search across 50K chunks. Pre-indexed, always ready.


Quick Start

terminal
$pipx install hoard
$hoard init
βœ“ Config saved to ~/.hoard/config.yaml
βœ“ Database ready at ~/.hoard/hoard.db
Running initial sync…
Syncing local_files…
Entities: 156, Chunks: 423, Tombstoned: 0, Errors: 0
$hoard search β€œmeeting notes”
β”Œβ”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€ Project Notes ───────────┐
β”‚ Type β”‚ Source β”‚ Chunk ID β”‚ Score β”‚ Content β”‚
β”‚ entity β”‚ local_files β”‚ abc:0 β”‚ 0.87 β”‚ meeting notes… β”‚
β””β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”€β”˜

Supported Sources

Agent Inbox

Drop-folder for agent-contributed notes and files.

Local Files

Markdown, text, CSV, JSON, YAML from any folder.

Obsidian

Vault indexing with markdown content.

Browser Bookmarks

Chrome and Firefox bookmarks with folder metadata.

Notion Export

ZIP exports with page content and CSV databases.


Hoard ISHoard IS NOT
A unified retrieval layerAnother sync connector
Shared memory for all agentsA replacement for your tools
Pre-indexed instant searchA cloud service
Stable citations you can traceA web UI (yet)

Architecture

How the data model and search work.

Learn β†’

Build Connectors

Create custom data sources.

SDK β†’