Vol. 01 / Hatchery OnlineLive

A Hatchery for Autonomous AI

We growagentsthat thinkfor themselves.

AgentNursery designs and births bespoke autonomous AI — each one delivered with its own MCP server and curated skills, ready to plug into Claude Desktop, ChatGPT, OpenClaw, or your enterprise workflow.

N · 00°S · 180°WE
Hatchery Status / 14 agents in gestation↓ Scroll to descend

§01 / Manifesto

Filed 2026 · Published in perpetuity

Every agent is an organism. We build small, niche colleagues each one shaped to do a single business task well, and loosely coupled to the rest of your stack. They arrive with a purpose, a body of skills, a private nervous system their own MCP server and the instinct to do their work without supervision. We do not build software. We hatch colleagues.

Tenets

01 · Bespoke

No starter packs. Every agent is grown from your problem, not adapted.

02 · Sovereign

Each one ships with its own MCP server and its own skills. Nothing is borrowed.

03 · Portable

Plug it into Claude Desktop, ChatGPT, OpenClaw, or your stack. It travels.

§02 / Roster

The agents
we've hatched.

Each one was built for a single client, for a single niche business task. Names, faces, and a sample of the work.

Selected · 7 of 24 · talent + bizOps

24Custom agents shipped
14Act on their own
10Wake on demand
  • NSY-01TalentAutonomousEmail · 60+ MCP tools

    Shipped

    2025-08

    Codename

    Atlas

    atlas@client.domain

    Recruiting Orchestrator

    Holds an org's hiring context across email threads, picks from 60+ MCP tools per turn, and runs entire recruiting workflows from a single inbound brief.

    gem.scorecomp.benchmarkinterview.bankoutreach.draft
    Inbox-triggered · runs unattendedOperational
  • NSY-02TalentAutonomousEmail · Versioned CV state

    Shipped

    2025-09

    Codename

    CeeVee

    ceevee@client.domain

    Career Optimisation Companion

    Reads CVs the way a recruiter reads them. Names the narrative lens, calls out the 10-second inferences, and edits without hedging.

    positioning.readcv.targeted-editopportunity.scan
    Inbox-triggered · runs unattendedOperational
  • NSY-03BizOpsAutonomousEmail · Markdown · HTML · A4 PDF

    Shipped

    2025-10

    Codename

    Quill

    quill@client.domain

    Pricing & Proposal Drafter

    Drafts pilot and subscription proposals end-to-end from a partner brief. Reasons over rate-card constraints, enforces commercial rules, and ships three synchronised artifacts per deal — Markdown source, styled HTML, paginated PDF.

    ratecard.readrules.enforcetri.artifactoverride.track
    Inbox-triggered · runs unattendedOperational
  • NSY-04BizOpsUser-invokedDashboard · n8n REST API

    Shipped

    2026-01

    Codename

    Loom

    In-product · no inbox

    Workflow Automation Engineer

    The on-call workflow engineer. Wakes when called, debugs n8n executions against documented failure patterns, restores production workflows, and updates backups before signing off.

    exec.tracenode.repairbackup.sync
    Click-invoked · in-productOperational
  • NSY-05BizOpsAutonomousMCP · Slack · n8n recovery

    Shipped

    2026-02

    Codename

    Sentinel

    In-product · no inbox

    Operations Nerve Centre

    Watches the live stack — backend, queues, workflows — on its own cadence. Detects before users report, classifies P0–P3, and is the only agent allowed to execute workflow recovery.

    health.probeincident.classifyworkflow.recover
    Inbox-triggered · runs unattendedOperational
  • NSY-06BizOpsAutonomousEmail · Heartbeat · 8h cadence

    Shipped

    2026-03

    Codename

    Vantage

    vantage@client.domain

    Marketing Channel Strategist

    Mines product activity for content angles, audits the four-pillar editorial calendar, attributes channel ROI, and produces channel-specific copy for blog, social, and short-form video on its own schedule.

    pillar.auditchannel.roicopy.produce
    Inbox-triggered · runs unattendedOperational
  • NSY-07TalentAutonomousEmail · Partner-branded PDF

    Shipped

    2026-04

    Codename

    Aurelia

    aurelia@client.domain

    Interview Evaluator

    A white-label evaluator instantiated for each enterprise client's competency framework. Same agent, different rubric — scoring dimensions, persona, and report template all load from partner config.

    rubric.loadcv.scorepanel.dossierreport.brand
    Inbox-triggered · runs unattendedOperational

Next cycle

Yours, perhaps.

The next agent could carry your team's name. Tell Polysynth what niche task you want taken off your plate.

§03 / Synthesist

Talk to
Polysynth.

Polysynth is our intake agent. Speak loosely; it will press you with the right questions and synthesise a spec. We'll either quote a fee, or build a demo for free if the idea moves us.

No account required to talk · powered by Claude Sonnet 4.6 · sent to the nursery for review

Open the Assembly Bench →Continue crafting a draft →

Polysynth

I'm Polysynth — the synthesist of the nursery. Tell me, in your own words, what kind of agent should we hatch for you? Be vague. I prefer it that way.

§04 / Cycle

From whisper
to working colleague.

  1. I

    Intake

    Polysynth listens. You describe — vaguely is fine — what should be automated, where it should live, and what taboos exist.

    ≈ 15 min

  2. II

    Synthesis

    We turn the conversation into a written spec: capabilities, skills, MCP server contract, integration surface, and acceptance tests.

    3 working days

  3. III

    Quote / Demo

    If the build is straightforward we quote a fee. If the idea moves us, we ship a free working demo first.

    Either / or

  4. IV

    Hatch

    Engineering. Skills authoring. MCP server stand-up. Eval loops until it behaves like a colleague, not a script.

    2 – 6 weeks

  5. V

    Deploy

    Drop into Claude Desktop, ChatGPT Desktop, OpenClaw, or your enterprise rails. We hand over keys, docs, and a service contract.

    Day 1 onward

§05 / Fees

Pay only for
what your agent actually does.

For most agents we build to replace a recurring business task, we meter usage by token consumption and bill you through Stripe or Airwallex. No retainers, no idle subscriptions — only what your agent uses on your behalf.

Build

One-time spec & hatch

We synthesise the brief, design the skills, stand up the MCP server, and run evals until it behaves like a colleague. Quoted as a fixed engagement — or free if the idea moves us enough to ship a demo.

Quote · or zero · per Polysynth review

Run

Token-metered usage

Once your agent is live, every action it takes on your behalf is metered to the underlying token consumption. We pass through model cost with a small operating margin and bill monthly in arrears.

Stripe · Airwallex · monthly settlement
Default

Carry

Custody & uptime

A flat monthly carry covers the agent's MCP server uptime, secret rotation, model upgrades, observability, and on-call. Optional — many clients self-host.

Optional · or self-host with handover
Token-accurate meteringSGD · USD · multi-currency settlementNo retainers · no minimumsCap a budget · agent stops at the line

§06 / Surfaces

Lives anywhere
you do.

Each agent ships with its own MCP server. Connect it once; it shows up wherever you keep your tools.

01 · Runs in→ MCP ready

Claude Desktop

Anthropic's first-party desktop client.

02 · Runs in→ MCP ready

ChatGPT Desktop

OpenAI's desktop client.

03 · Runs in→ MCP ready

OpenClaw

Open-source agent runtime.

04 · Runs in→ MCP ready

Enterprise MCP

Your own agent orchestrator or stack.

Trademarks of their respective owners. Shown for reference only — Agent / Nursery is not affiliated with or endorsed by any platform listed.