Skip to content

swarm#

Ship regulated ML faster. An opinionated multi-agent platform for training, deploying, and auditing machine-learning models in regulated industries — built for BFSI, healthcare, and EU-AI-Act high-risk workloads.

Quickstart — 5 minutes Concepts Tutorials

Why swarm#

  • Regulator-ready by default

    RBI FREE-AI compliance bundle, HIPAA-aligned audit trail, EU AI Act Annex IV technical documentation. Every agent action — tool calls, permission denials, model promotions — lands in a tamper-evident audit log.

  • Claude Code marketplace plugins

    Install any CC marketplace plugin unchanged: skills, commands, sub-agents, shell-command hooks, MCP servers. Tested against 16 real plugins across Anthropic's official + opensource marketplaces.

  • Composable, not framework-locked

    40 agents across 7 specialized teams, swappable backends (native / LangGraph), 38 auditable tools, 18 algorithm repos. Your pipeline is YAML, your agent is a dataclass, your tool is a Python function.

  • Unified permission engine

    RBAC, per-agent tool allowlists, feature flags, HITL approval gates, and YAML policy all resolve through one ALLOW > DENY > ASK pipeline. Every denial attributes the source rule — SQL-queryable, BFSI-auditor-ready.

  • Champion-challenger + shadow traffic

    Promote models with guardrails: shadow-traffic comparison, fairness delta, drift detection. Audit PDF generated on every promotion.

  • Scheduled ops, not cron-by-hand

    Retraining, nightly drift checks, audit-PDF rollups, retention sweeps — all first-class cron primitives with REST + CLI + dashboard surfaces.

Who it's for#

You are… swarm gives you…
An ML engineer at a BFSI mid-market shop Production pipelines with regulator-ready audit trails in one afternoon instead of a 2-week compliance ticket
A CRO / Head of AI risk A compliance-profile switch (RBI / HIPAA / EU AI Act) that produces defensible evidence in a format your regulator already accepts
A platform engineer Kubernetes-native deployment (Helm), OIDC SSO, BYOK, multi-region, retention automation
A data scientist Clean Python SDK, Jupyter-first tutorials, declarative pipelines, fast iteration on a laptop
A plugin author Anthropic's Claude Code format works unchanged — skills, commands, sub-agents, hooks

Ready?#

  • 5 minutes, laptop — Quickstart with docker compose up
  • 30 minutes — Walkthrough — fraud classifier end-to-end
  • Hands-on demo with us — email hello@theaisingularity.org