Curated agent memory for the whole team.

Instructions, memories, facts, preferences, contacts. Everything a new hire learns over time, your agents start with out of the box. Curated by humans, scoped by team, served to any agent over MCP.

Create your workspaceRead the docs
how memory reaches your agents● live
Your team writes it down
MUST rulesSHOULDGlossaryContacts
CURATE · APPROVE · VERSION↓
Lore IndexGOVERNED CORE
scopedversionedaudited
Compiled bootstrap pack · caps context at ~1,500 tokens
ONE MCP ENDPOINT↓
Claude
via MCP
Cursor
via MCP
Your agent
via MCP
01

Governed

Every entry is human-curated, scoped, versioned, and audited. Agents can propose new entries — proposals are approved by humans in your organization, not AI.

02

Vendor-neutral

One MCP endpoint compatible with all MCP-enabled agents. Work with Claude, Cursor, or your own.

03

Fight context bloat

Every agent session starts with a bootstrap pack under 1,500 tokens — only the most important entries, nothing more. Agents search for what they need after that.

How it works

Write it down once for every agent in your organization.

1

Write

Capture rules, preferences, facts, glossary, contacts, and anything else agents need to know about your organization.

2

Scope

Assign each entry to a team, project, or the whole org.

3

Version

Humans approve changes. Every edit is versioned and audited.

4

Serve

Agents start with a narrow bootstrap pack of critical entries, then search for more only when they need it.

Set up any agent with Lore Index and watch it join the team.

Connect your agent to Lore Index via our MCP server. It starts every session with the bootstrap pack, and can search for any relevant information it needs later on.

  • ◆ Compatible with all MCP-enabled agents
  • ◆ Task-aware search keeps agent context light and relevant
~/.config/mcp/lore.json
// Add Lore Index to any MCP-capable agent
{
  "mcpServers": {
    "lore": {
      "url": "https://api.loreindex.ca/mcp",
      "headers": {
        "Authorization": "Bearer lore_sk_..."
      }
    }
  }
}
→ bootstrap loaded · 1,480 tokens · 12 MUST · 34 SHOULD
The difference

Extraction guesses. Governance decides.

Memory by extraction
  • ✕ Scraped from chats — nobody approved it
  • ✕ Silos per vendor, drifts out of date
  • ✕ Balloons context, hard to audit
  • ✕ No way to say "this is binding"
Memory by governance
  • ✓ Human-curated, approved, and cited
  • ✓ One vendor-neutral MCP endpoint
  • ✓ ~1,500-token compiled bootstrap
  • ✓ MUST rules are binding by design

Stop treating every agent like a new hire.

Write it down once, scope it to teams, and create the perfect memory database for your organization's agents.

Create your workspaceSee pricing