Skip to content
BetaCited search for small and midsize teams

Know exactly what you promised. Source included.

Leon syncs your meeting notes, tickets, CRM, support inbox, email and Slack, plus your team’s AI conversations. Ask in plain English and get an answer where every sentence links to the original. If nothing on record supports an answer, Leon says so.

Open beta · no billing yet

Fictional example.

Question: what did we promise ACME for V2?

Leon’s answer: At the Oct 3 monthly check-in, Julie committed to delivering V2 on Nov 15 [1]. The signed quote confirms the date but excludes the data migration [2], and the migration ticket is still open [3].

  1. Meeting, ACME monthly check-in, Julie Martin, Oct 3, 2026
  2. Note, Signed V2 quote, Karim Benali, Oct 7, 2026
  3. Ticket, ACME: data migration, Sophie Laurent, Oct 21, 2026

Connectors

42 connectors, plus your AI assistants.

Meetings, tickets, CRM, support, docs, email and Slack. Leon pulls changes every 15 minutes over OAuth or an API key.

  • Dev5
  • Meetings8
  • Project management13
  • CRM5
  • Support8
  • Email, chat and calendar3
See all 42 connectors
  • GitHub
  • Notion
  • Linear
  • GitLab
  • Sentry
  • Granola
  • Fathom
  • Fireflies
  • tl;dv
  • Circleback
  • Claap
  • Leexi
  • Gong
  • Jira
  • Confluence
  • Asana
  • ClickUp
  • Trello
  • monday.com
  • Shortcut
  • Plane
  • Airtable
  • Basecamp
  • Miro
  • Productboard
  • Google Drive
  • HubSpot
  • Attio
  • folk
  • Pipedrive
  • Close
  • Intercom
  • Zendesk
  • Freshdesk
  • Help Scout
  • Front
  • Plain
  • Crisp
  • Missive
  • Outlook
  • Gmail
  • Slack

The problem

Your team already wrote the answer down, in one of a dozen tools.

  • Context is fragmented

    Email, CRM, tickets, meeting notes, docs. Finding one commitment means searching each tool separately and hoping you picked the right one.

  • Context walks out the door

    When a project manager leaves, their commitments leave with them. Your client remembers every one.

  • Decisions happen in private AI chats

    People work through decisions with AI assistants, and those threads stay buried in each person’s history. Leon indexes them, privately, alongside everything else.

Fictional examples

How it works

Connect your tools once, then ask Leon what they contain.

Leon syncs every 15 minutes and stores each item verbatim. Your team keeps working in the tools it already uses.

  1. 01

    Connect your tools

    OAuth or an API key, across 42 connectors. The browser extension and a coding-assistant hook bring in AI conversations.

  2. 02

    Leon indexes the originals

    Stored verbatim, with a deep link back to the source. AI conversations stay private to their author. Credentials and tokens are redacted before storage.

  3. 03

    Ask

    From search, chat, or your AI assistant over MCP. Each sentence of the answer links to its source.

Fictional examples

Architecture

Every query passes through three layers, in this order.

Leon never rewrites your data. It stores the originals, indexes them for hybrid search, and enforces permissions before anything is ranked.

  1. 1Source of recordEvery meeting, ticket, note and email is stored verbatim, with a deep link to the original.
  2. 2Search indexChunked into passages and indexed twice, Postgres full-text and vector embeddings. Fuzzy matching on titles absorbs typos.
  3. 3Access controlRow-level security filters every query before ranking. Private content never reaches anyone else’s results.

Fictional example

Use cases

Six teams search the same index for different answers.

Leadership, sales, support, product, project managers and new hires search the same tools for different things.

Know exactly what you promised. Source included.

A meeting three months ago, a signed quote, an email thread. When a client disputes scope, Leon retrieves the record and cites it. When there is no record, it says so.

Typical queries

  • what did we promise this client in last month’s meeting?
  • the client says it was out of scope: what did we write?
  • what was the project manager who just left working on?
  • type:meeting since:7d

Team tools: HubSpot, Outlook, Granola, Notion, Asana, Fathom

See the full use case
what did we promise ACME for V2?
Leon’s answer3 sources

At the Oct 3 monthly check-in, Julie committed to delivering V2 on Nov 15 source 1. The signed quote confirms the date but excludes the data migration source 2, and the migration ticket is still open source 3.

  1. [1]ACME monthly check-in

    Meeting · Julie Martin · Oct 3, 2026

    “We can commit to November 15 for V2, history migration excluded.”
  2. [2]Signed V2 quote

    Note · Karim Benali · Oct 7, 2026

    Delivery on 11/15/2026. The data migration is quoted separately.

Fictional example

Design principle

Every sentence
comes with its source.

An unsourced answer is worthless the moment a client disputes it. Leon shows where every sentence comes from, and stops when the record runs out.

what price did we give Verlant in 2024?

I don’t know. No source on record mentions a price agreed with Verlant in 2024.

Fictional example

  • Sentence-level citations

    Every claim links to the meeting, ticket or email it comes from.

  • “I don’t know” when the record is silent

    When the sources don’t answer the question, Leon says so instead of filling the gap.

  • Permission-aware everywhere

    Search, chat and MCP enforce the same access rules. Private content never surfaces for anyone else.

  • The original text, opened in its own tool

    Leon stores what was written verbatim, and every citation opens the meeting, ticket or email where it was written.

MCP server

Give your AI assistant your team’s context.

Leon’s MCP server exposes four tools to any MCP client: search, fetch a source, get the morning briefing, save a note. It authenticates with a revocable personal token and inherits your permissions exactly.

  • Your coding assistant opens each session with the morning briefing.
  • Coding sessions are captured when they end, text only.
  • Each source your assistant opens in full is written to an audit log.
Chat, morning briefing and MCP
coding assistant · Leon MCP server
Morning briefing08:45 · 250 words max

What you did

  • Reviewed and merged the PDF export pagination PR.
  • Replied to ACME about the V2 timeline.

What changed

  • Sophie opened the ACME data migration ticket.
  • Two new support conversations mention the export bug.

Still open

  • The revised Verlant quote, promised last week.

Fictional example

Security

Each company’s data is isolated in Postgres, and AI conversations stay with their author.

More on security
  • Tenant isolation

    Each organization’s data is isolated at the database level with Postgres row-level security, and covered by automated tests.

  • Permission-aware retrieval

    Search, chat and the MCP server run with the permissions of the person asking. Access is filtered before ranking, so hidden content never shapes a result.

  • Private conversations

    AI assistant conversations, chats with Leon and saved notes are visible only to their author.

  • Encrypted credentials

    API keys and OAuth tokens are encrypted with AES-256-GCM and bound to their database row.

We also publish what’s missing: see the list.

FAQ

Common questions.

Pricing and FAQ
Do we need to switch tools?

No, as long as they are among the 42 supported connectors. Leon connects over OAuth or an API key and syncs changes every 15 minutes.

Who can see what?

Content from a connection is shared with the whole team, as in a shared tool, except email (Outlook, Gmail), Google Drive and Slack, which stay private to whoever connects them. AI assistant conversations, chats and notes are visible only to their author.

What happens when Leon doesn’t know?

It says so. Every claim in an answer is cited. Without a source, the answer is “I don’t know.”

Does Leon summarize our documents?

No. It stores the original text verbatim and indexes it for search. Each claim in an answer cites that text, with a link to the original.

Everything your team said, on record.

Create a workspace, connect your first tool, and ask your first question.