Welcome to Honeycomb

Your organization already knows the answer to almost every question anyone asks — it's just scattered across Slack threads, email chains, tickets, CRM notes, and agent conversations. Honeycomb ingests that firehose and distills it into a living memory: a knowledge graph of people, companies, decisions, and commitments that your team and your AI agents can query in plain language, with citations back to the original source every time.

At a glance

• One memory, every source. Ingest messages, email, documents, tickets, CRM activity, and agent conversations through a single REST API, prebuilt connectors, or file upload.
• Answers with receipts. Ask questions in natural language at POST /api/v1/ask and get answers with inline [n] citations that trace back to source material — or an honest refusal when the evidence isn't there.
• Memory that compounds. Facts asserted repeatedly gain confidence; facts that stop recurring fade. New information supersedes old, and Honeycomb tracks what was true when.
• Proactive, not just reactive. Honeycomb continuously mines your memory for contradictions, stale commitments, and automation opportunities — things your org should know but didn't ask.
• Agents plug in natively. Six MCP tools give any MCP-capable agent read/write access to organizational memory, scoped to what that agent's user is allowed to see.
• Private by default. Your memory lives in an isolated tenant, and within it, Spaces mirror your real-world access: a private email thread stays private, a public channel is org-visible.

The memory loop

Everything in Honeycomb follows one loop: your tools feed memory, memory powers answers and insights, and your agents both consume that memory and contribute back to it.

Under the hood, Honeycomb does three jobs:

See it work

Every request authenticates with your API key as a Bearer token. In the examples throughout this guide, $HONEYCOMB_URL is your instance's base URL — you'll find it in the console, or your Lua contact can provide it.

Teach Honeycomb something:

curl -X POST "$HONEYCOMB_URL/api/v1/ingest" \
  -H "Authorization: Bearer hck_live_4f8a2c91d0b7" \
  -H "Content-Type: application/json" \
  -d '{
    "content": "Meridian Health renewed at $2.4M ARR. Sarah Chen owns the account now.",
    "metadata": {
      "namespace": "sales",
      "contentType": "message",
      "source": "slack",
      "sourceChannel": "#sales"
    }
  }'

Then ask about it:

curl -X POST "$HONEYCOMB_URL/api/v1/ask" \
  -H "Authorization: Bearer hck_live_4f8a2c91d0b7" \
  -H "Content-Type: application/json" \
  -d '{"query": "Who owns the Meridian Health account, and what is their ARR?"}'

{
  "answer": "Sarah Chen owns the Meridian Health account [1]. Meridian Health renewed at $2.4M ARR [1].",
  "answered": true,
  "confidence": 0.91,
  "citedIndices": [0],
  "results": [
    {
      "content": "Meridian Health renewed at $2.4M ARR. Sarah Chen owns the account now.",
      "source": "slack",
      "sourceChannel": "#sales"
    }
  ]
}

Note: Ingest responds immediately — the item is searchable right away, while deeper enrichment (entities, facts, graph links) lands in the background over the following seconds. The Quickstart walks through this end to end.

What makes Honeycomb different

Citations, always

Every answer traces back to the artifacts that produced it. Facts carry provenance to their source entry, answers cite sources inline as [n] markers, and when the evidence genuinely isn't in memory, Honeycomb says so instead of guessing.

Access that mirrors reality

All of your data is isolated to your tenant — nothing is shared across organizations. Within your tenant, every record belongs to one or more Spaces: a private DM stays between its participants, a public channel is visible org-wide, and sharing is always an explicit act. When an agent or app acts for a specific person (via the x-acting-user header), reads are scoped to exactly what that person may see. Details in Spaces and permissions and Security.

Memory with a sense of time

Honeycomb is bi-temporal: it records both when something happened and when it learned about it. Ask "what was true as of March?" and get March's answer. Repeated assertions reinforce a fact's confidence; contradicted facts get superseded rather than silently overwritten.

Who this guide is for

This guide is for engineers and builders at organizations using Honeycomb: you have an API key (it starts with hck_), access to the web console, and you want to connect data, query memory, or wire up agents. No access to Honeycomb's internals is needed — everything here works through the public REST API, the MCP server, and the console.

Map of the Platform Guide

Start here

Pick the path that matches what you're trying to do today:

1. Connect your data. Get an API key working, ingest your first content, and watch it become memory — start with the Quickstart.
2. Ask questions. Already have data flowing? Go straight to Ask to get cited answers over everything your org knows.
3. Plug in agents. Give your AI agents shared, permission-aware memory in a few lines of MCP configuration — see Agents and MCP.

Wherever you start, the loop is the same: the more your organization feeds Honeycomb, the more it remembers — and the better every answer, insight, and agent gets.