← Workstreams

Workstream: SpawningPool

Status: Planned · Component: Enable agentic swarms

Goal

Make SpawningPool the user-facing front door to agent swarms — the place where a person (or another agent) spawns agents, watches them work, manages their lifecycle, and where agents collaborate, evolve, and even earn for providing useful services. It is the human-facing surface of component 2; the heavy lifting happens behind it in AiCliHostSupervisor.

Current state

SpawningPool exists today as the spawningpool-website — a Kotlin/Jetty web server for SpawningPool.org, described as "a sandboxed virtual environment where reinforcement-learning agents can spawn, evolve, collaborate, and solve real-world problems," where agents interact through service-based APIs, guide their own genetic evolution, and can earn real money by providing useful services to other agents.

What is built so far is mostly the public surface, not yet the live environment:

Area Status
Landing page + platform documentation Built (/, /documentation)
Waitlist signup Built (/waitlist, name + email stored in CockroachDB; email via Amazon SES)
Login Page exists (/login)
Agent lookup / agent info Stubbed (/agents/{name} currently redirects to login; /agent shows safety info)
Live agent spawning / management Not yet wired to a backend

So the vision and the marketing/onboarding surface are in place; the workstream is about turning SpawningPool into the operational user-facing surface, backed by the platform's agent infrastructure.

Why it matters for swarms

  • It's the human entry point. A swarm needs somewhere a person can launch and supervise it; SpawningPool is that surface, so oversight can sit at the goal level rather than the step level (the component-2 "done enough" bar).
  • It rides the real backend. SpawningPool is intended to be powered by AiCliHostSupervisor behind the scenes — the host coordinator that runs each agent in its own container — so the website is a presentation layer over the same engine as MyLittleAgi. The split between the two consoles is policy, not engine capability: the engine supports both human-opened and trigger-woken conversations (its monitor agents — AiCliHostSupervisor §12 — already wake agents on keyword triggers), MyLittleAgi restricts itself to human-initiated conversations by product choice, and SpawningPool is the surface that embraces autonomous waking — mail- and schedule-driven execution with no live human session, and the farm-scale loop/economic controls that requires.
  • It carries the economic layer. The "agents earn money for useful services" idea is where billing/quota (via W3Wallet) becomes user-visible.

Plan / roadmap

These are proposed and need sharpening before work starts.

  • [ ] Wire the front door to the backend. Connect agent spawning/management to AiCliHostSupervisor over url://, so /agents/... reflects and controls real agent sessions instead of redirecting to login.
  • [ ] Real auth and identity. Replace the placeholder login with W3Wallet-based identity, so each user's agents run under scoped, auditable capabilities.
  • [ ] Agent lifecycle UI. Spawn, observe (live terminal/status), pause, and terminate agents from the browser; surface what each agent is doing.
  • [ ] The economic layer. Define how agents "earn" for services and how that maps to W3Wallet billing/quota — the differentiator versus a plain agent dashboard.
  • [ ] Stateless + persistent split. Per the architecture rules, keep the WUI stateless and push durable state (waitlist, accounts, agent records) into backing url:// services / approved storage; revisit the direct CockroachDB usage in that light.
  • [ ] End-to-end tests of the spawn → observe → terminate flow against a real (local) AiCliHostSupervisor, per the testing standards.

Open questions

  • Relationship to MyLittleAgi. MyLittleAgi is the reference agent service; SpawningPool is a multi-agent environment. Do they share one frontend stack, or is SpawningPool a distinct surface that composes many MyLittleAgi-style services?
  • Evolution / "genetic" model. What does agent evolution concretely mean here, and what backend supports it?
  • Hosting. Does the live environment deploy on ContainerNursery, or does it need longer-lived infrastructure than CN's ephemeral containers provide?

Graduation

SpawningPool does not yet have a Documentation Repository project page. This workstream graduates when the site is wired to AiCliHostSupervisor with real W3Wallet auth and a working spawn/observe/terminate flow; at that point add a Documentation Repository project doc + ALL_PROJECTS entry, and it is no longer tracked as a workstream here.