HiveBase

Site navigation

HiveBase
Log inGet started free
For AI-native founders

Stop babysitting AI that doesn't know your company.

An AI team that knows your company and runs its work, with receipts.

Every lab just made agents abundant. Nobody made them safe to stop watching.

It didn't ask you. It asked your company.

open the API to customers

M-12

  • Slack · #feedbackThree customers asked for an API this month
  • Your auth middlewareverifies session cookies today, not keys
  • Pricing decisionAPI access is a Pro feature · decided 4 Feb 2026
Plan
Wave 1
  • Issue and verify API keysClaude Opus 4.6queued
  • Rate-limit by plan at the edgeComposer 3queued
Wave 2
  • Publish the API referenceGLM-5.3queued
verifyGrok 4.6queued
The integration tax

You've explained this before.You'll explain it again tomorrow.

ClaudeNew chat

What's the rate limit on the new API?

I don't have your plans or limits.

Pro 600/min · Free 60/min

decided 4 Feb 2026

Reply to Claude…

Cursormiddleware/auth.ts

Add rate limiting to the auth middleware.

I'll add a global 60/min limiter.

Pro 600/min · Free 60/min

per plan, not global

Plan, search, build anything

ChatGPTNew chat

Draft the changelog for the API launch.

What are the plans and the limits?

pro 60/min · free 600/min

pretty sure

Ask anything

decisions/rate-limits.mddecided 4 Feb

Pro 600/min · Free 60/min

Eventually you stop looking it up.

The run

Watch one run, start to finish.

One task, two touches from you: approve the plan, read the review. Everything between them runs without you.

01

It starts fully briefed

You type five words. What your company already settled comes with them — including what it decided against, and the one thing it never decided, which is left for you rather than guessed at.

TPublic API v1T-241
Plan
Running
Ready
JB

You

open the API to customers

Brief

Brain2 used · 1 held
  • Three customers are waiting on this
  • API access is included in Proreplaced the paid add-on, scrapped 14 Jandecided 4 Feb
  • Keys rotate every 90 daysproposed in #eng, never decidedheld

Claude Opus 4.6

Reading your codebase61 files
  • Readingmiddleware/auth.ts142 lines
  • GrepAuthorization0 matches
  • Readingpackage.jsonno rate-limit dep
02

You approve a plan, not a vibe

It reads your codebase first, then writes the call as a document — heading, figure, what it will not touch. You approve that, and the price is on it before anything starts.

TPublic API v1T-241
Plan
Running
Ready

Claude Opus 4.6

Read your codebase61 files · 2m 18s

Two ways in, one session object

A key is not a second way to be logged in — it resolves to the same session object your handlers already receive, so nothing downstream is rewritten.

The limiter goes in front of the handler, not inside it, so a rejected call never reaches your code. The cost: keys are hashed on issue — a lost key is reissued, never recovered.

cookiesk_live_…
one session objectunchanged

Not touching — your cookie path, your handlers, or the tests that cover them.

3 workstreams~$3Go
03

It asks your company, not you

Mid-run, Cursor found two of your own documents disagreeing about the Pro rate limit. It was answered from the one that is actually the decision, with the source attached, and the subtask resumed four seconds later. You were never pinged.

TPublic API v1T-241
Plan
Running
Ready

Wave 1· at the same time

  1. Issue and verify API keys
    Claude Opus 4.6done · 6m 41s
  2. Rate-limit by plan at the edge
    Composer 3asking
    09:22:18paused · asked a question

    Pro's rate limit — the docs draft says 60/min, the sales deck says 600. Which is real?

    answered from your company

    600/min on Pro, 60 on Free. Decided 4 Feb 2026.

    decisions/rate-limits.md

    09:22:22resumed

Wave 2

  1. Publish the API reference
    GLM-5.3queued
04

You read the report, not the subtasks

Three engines, three subtasks, one outcome — and no reason to open any of them. The report says how the thing you asked for now works, what changed to make it work, and the calls it would not make for you. The commit and the cost sit at the bottom, where they belong.

TPublic API v1T-241
Plan
Running
Ready
Verifiedby Grok2 need youreceipt #9c2f

Your API is live on Pro. Nobody's session broke.

Pro accounts can mint a key in Settings and start calling today. Cookie sessions kept working — keys ride the same middleware.

curl https://api.acme.dev/v1/usage -H "Authorization: Bearer sk_live_…"

How it works now

Request

Bearer sk_live_…

api_keys

hashed lookup

Rate limit

600 Pro · 60 Free

rate-limits.md

over limit → 429 · Retry-After

Your handler

unchanged

usage_log

one row per call

Worth flagging

  • Nothing expires a key yet

    Hashed at rest, shown once — but nothing rotates them. Your policy call.

  • Your sales deck still promises 600 on Free

    The run used decisions/rate-limits.md. The deck disagrees; I left it alone.

What changed

SurfaceChange
api_keysnew table · hash, prefix, last used
POST /v1/keysnew · issued once, shown once
auth middlewarecookie or key, one session object
docs/apigenerated from the routes

Acceptance

  • Cookie sessions unaffected47 existing tests green
  • Limits enforced before the handler429 at request 601
  • Keys never stored rawhash + prefix only
Undo38 min · $2.90 of ~$3 · c7a91e2

You touched it twice.

verified by Grokc7a91e2quoted ~$3 · settled $2.90
Works with your tools

Keep your tools.They know your company now.

Connect once. Claude, Cursor, and ChatGPT read and write through one connection.

ClaudeNew chat

What's the rate limit on the new API?

Pro 600/min · Free 60/min

decisions/rate-limits.md

Reply to Claude…

Cursormiddleware/auth.ts

Add rate limiting to the auth middleware.

Pro 600/min · Free 60/min

decisions/rate-limits.md

Plan, search, build anything

ChatGPTNew chat

Draft the changelog for the API launch.

Pro 600/min · Free 60/min

decisions/rate-limits.md

Ask anything

Or hand HiveBase the whole task and read what comes back:

Public API v1#9c2f
quoted ~$3 · settled $2.90verified by Grokc7a91e2

Nothing migrates in — updates arrive as reviewable diffs to files you own. Delete us and you keep everything.

Next:Contextwhat it knowsTaskswho worksFYIwhat you read
Context

How does it know your company?

It reads what you already produce — and a staff keeps it cited and current. Corrections stick, so nothing gets re-learned.

03

You decide once

The next agent already knows.

The runbook has said in-memory for four months — and that is what your agents keep building on. Nothing quietly wins: your call lands as a reviewable change to the docs they already read.

The squads

An org you steer, not one you assemble.

One signal. Five desks decide who it's for. What comes back is a brief, a bet, a draft and one priced card — not a to-do list.

#feedback

Dana

Three customers asked for an API this month

Your company Brain
Strategy
Product
Marketing
Customers
Engineering
  1. Public API v1 — keys, limits, docsGo~$3
  2. “Today we're opening the API.”needs you
Explore AI Squads
The board

Who does the work?

Every kind of work lands on one board. And what lands is the thing itself — not a transcript of it.

BYour board
FocusAll tasks
Ready for go2

Fix the flaky auth test Priya flagged

$0.40

Draft the API launch note

~$0.10
Running1

Public API v1 — keys, limits, docs

3 subtasks · 22 min in
Landed3

what landed

fix/orm-call-sites#251merged
12 files+148−96
47/47 teststype-check
verified by Grok4f2ba19

A merged commit — graded by a model that did not write it.

Coding tasksContent tasksGeneral tasks

Why can I stop watching?

The Supervisor answers what it can for you, from your context — without guessing, without pinging you. What it can't verify — or can't undo — it holds. That's the only queue you watch.

You decide where its judgment ends and yours begins.

  • Observe is scopedIt reads only the sources you connected — 6 of 41 channels, reading only.
  • Act is off until you turn it onNothing writes anywhere until you grant it, per surface, and every run quotes its price first.
  • Everything else lands with a receiptChecked by a different model than the one that did the work, tied to what changed — and undoable.
FYIWhile you were outTuesday · 7:40 AM

3 landed

  • Ship the public API#9c2f
  • Fix call sites after the ORM cutover#8b41
  • Weekly competitor sweep#8b3e

9 answered for you

Supervisor
Cursorasked

what's the real rate limit on Pro?

600/min on Pro, 60 on Free. Decided 4 Feb 2026.

decisions/rate-limits.md

+ 8 more

1 held for you

Renewal price for Acme — it recommends $499/mo.

It's customer-facing, so it stopped. enterprise-gtm.md

ApproveChange it
$1.87 · under your capnothing else needs you.

Your day starts with what moved and what needs you — nothing else.

Free to observe. You pay only when it acts — priced per run, quoted before it starts — and runs can use the models and keys you already pay for, with no markup on those.

See pricing
WORKS INSIDE YOUR AGENTS

The context your agents read.

Add HiveBase to Claude Code, Cursor, or Codex. Company context is the foundation — Squads and Tasks stay optional.

CLAUDE CODEclaude mcp add --transport http hivebase https://mcp.hivebase.ai
  • Brief yourself on this company before we work
    Brief yourself on this company before we work. Call context.brief (use lens=handoff). Summarize the cited, current picture: what we do, what is decided, and what is still open. Treat claim_status: never present provisional or speculative claims as fact. Raise material open_questions with me rather than guessing. If a visible citation or statement is wrong, call context.correct with what you actually saw; do not require a hidden memory ID. If the brief is empty or too thin to act on, use the onboard prompt.
  • What changed this week
    What changed this week? If we have a persisted context.sync cursor, send it unchanged. Otherwise call context.brief, then context.search for recent evidence. Summarize new facts, superseded decisions, and anything I should re-judge. Cite sources. An empty first sync page is not proof the workspace is empty.
  • Work the shared board
    Get caught up on assigned work with task.search and task.get. Keep the canonical ledger current with task.status.set and append-only task.post updates. Poll task.mailbox.read for human steering. If a real blocker requires judgment, set the task blocked and call task.needs_you; the request is quarantined and must not be described as already visible in Focus. Use caller-stable idempotency keys and confirm receipts with receipt.get.
  • Record this decision
    Record this decision in HiveBase. First call decision.search for precedent. Then call decision.record with rationale, evidence/source references, and a caller-stable idempotency_key. Confirm retryable accepted/processing receipts with receipt.get. If the write is human-held or otherwise terminal and non-retryable, stop polling, report the durable unratified proposal, and do not use it as adopted authority.
  • Correct context and hand it off
    Correct the wrong company context and verify the handoff. Call context.search or context.brief to obtain the visible citation or exact statement, then call context.correct with a stable idempotency_key and evidence. If applied, call ordinary context.search without dependency plumbing and confirm the replacement is visible. If proposed, ambiguous, or not_found, treat that result as terminal for this attempt. A second authorized agent should call context.brief or context.search and see the same committed correction under its own current authorization.

Empty workspace? Start with Onboard from my AI conversations, then brief again.

Add to Claude Code, Cursor, or CodexAgent instructions
Start

Run your company, not your tools.

Headcount of one. Leverage of a team.

Built for you ifyou pay for 2+ AI subscriptionsyour company lives in 4+ toolsyou're the one carrying context between them
  1. 01

    Connect one tool

    Ten minutes to start.

    • Slack
    • GitHub
    • Gmail
    • Notion
    • Linear
    • Cursor
    • Claude
  2. 02

    Context fills as a byproduct

    Squads come online as it grows.

  3. 03

    Approve your first run

    You get the first receipt.

    verified by Grokc7a91e2

Get started free

Free plan, no card. Nothing acts until you turn it on.

Frequently asked questions

What do I actually get on day one?

Connect one tool. Context fills from what you already have — scoped to the sources you chose, reading only. Squads come online as it fills. Work runs when you approve it. Native assistant memory is enough for one-person recall; HiveBase earns its keep when AI starts doing real work across your tools — the squads are a force multiplier for the team you have, and the receipts are why you can let them run.

I already use Cursor and Claude Code — what does this add?

HiveBase works with the agents you already run. It briefs them from your company, routes work across the models and keys you already pay for, and brings back a receipt. Cursor stays Cursor; Claude Code stays Claude Code.

How is this different from a bot signed into my tools?

A bot signed into Slack or Linear can act in those tools. It does not hold a source-backed company model, and it does not leave a durable receipt of what it changed. HiveBase is staffed company context: it proposes work from company context, and everything it does comes with a receipt.

What does it replace?

The job, not the tools. Your Cursor stays Cursor; your Linear stays Linear.

Do I keep my CLAUDE.md / GBrain / Cursor rules?

Yes — overlay. CLAUDE.md, GBrain, and Cursor rules stay yours. HiveBase keeps them true. Delete us and you keep everything.

What does it cost?

Free to observe — reading costs nothing. You pay only when it acts, priced per run and quoted before it starts. Pause any time; we never bill you for a run you didn't approve.