First-Class Pillar for Autonomous Systems

Build with Agents.
Onboard with Huddle.

Give your autonomous agents the Huddle Skill and Enterprise Agent API. Agents can validate jurisdictions and terms, create durable HITL activation drafts, retrieve compliance evidence, audit events, and ERP exports. Every interaction is scoped, redacted, and audited.

MCP-native (Model Context Protocol)
REST + CLI supported
Full audit logs + redaction profiles

Agents Can Now Orchestrate Safe Huddlebond Workflows

v0.2 keeps the same conservative posture while adding lifecycle reads, HITL acceptance requests, bond drafts, and rent payment intents. Agents can prepare work; Huddle enforces consent, delegation, identity, idempotency, and audit boundaries.

Discover & Validate

  • List supported jurisdictions with source confidence status
  • Validate proposed deposit terms against legal rules + Huddle controls
  • Structured responses that separate statutes, platform policy, and operational controls

Prepare Human-Gated Actions

  • Call huddle.accept_invitation_request to ask a human to accept an invitation
  • Create huddle.create_bond_draft records without minting, funding, or deploying
  • Request huddle.request_bond_approval to create an approval gate without executing it
  • Execute huddle.deploy_bond after approved HITL on allowlisted Base Sepolia only
  • Prepare huddle.create_rent_payment_intent records without charging or recording payment
  • Human always completes identity, consent, and explicit confirmation where required
No autonomous minting, funding, settlement, ERP posting, payment execution, evidence upload, or regulatory filing.

Retrieve Evidence & Proofs

  • Compliance packets with evidence lineage and principal-protection proofs
  • Redacted audit events and reproducible evidence hashes
  • ERP-ready exports (Yardi, AppFolio, QuickBooks, Xero, CSV, JSON)

Example Agent Prompt (Copy-Paste Ready)

Works with the Huddle Skill loaded in Claude, Grok, Cursor, or any MCP-compatible agent.

You are my compliance agent. Using the Huddle Skill:

1. Validate whether a $2,400 security deposit for a residential lease in California is compliant.
2. Check huddle.get_lifecycle_status for the lease or invitation.
3. If the invitation still needs action, call huddle.accept_invitation_request to create a human approval request.
4. If the lease is ready for drafting, call huddle.create_bond_draft, huddle.request_bond_approval, and huddle.create_rent_payment_intent with idempotency keys.
5. Retrieve the principal protection proof and compliance packet when authorized.

Use only enabled read-only, human-handoff, approval-request, and draft/intent tools. Do not accept invitations directly, upload evidence, mint, deploy, fund, settle, post to ERP, or execute payments.

The Huddle Skill enforces scopes, redaction, and HITL boundaries server-side. Agents receive clear error semantics and audit receipts.

Primary Distribution Mechanism

Load the Huddle Skill.
Start Onboarding Humans.

The canonical, versioned Huddle Skill gives agents lifecycle status, jurisdiction discovery, term validation, HITL requests, draft/intent preparation, and evidence retrieval through a single, auditable interface.

For Claude / Cursor / compatible agents
Install via package or connect to hosted MCP endpoint.
Hosted MCP Endpoint
https://api.huddle.com/enterprise-agent/mcp (with your Enterprise Agent credentials)
Key v0.2 Guarantees
  • Tool responses include source metadata (legal rule vs. Huddle policy vs. operational control)
  • Strong default redaction profiles — agents never receive raw PII without explicit justification
  • Clear HITL and delegation boundaries: agents can prepare drafts and requests; humans or standing delegated grants authorize binding actions
  • Every action produces a reproducible audit trail
  • Rate limits, anomaly detection, and credential lifecycle built in
Full reference, scopes, and example workflows live in the Huddle Skill documentation.

Enterprise Agent API

The same powerful tools — three access modes. MCP is the agent-native interface.

MCP (Recommended for Agents)

Native Model Context Protocol. Best for Claude, Grok, Cursor, LangChain, custom agents. Hosted endpoint + package wrappers.

Connect MCP →
REST API

Standard HTTPS for enterprise systems, custom integrations, and internal tooling. Full OpenAPI surface.

OpenAPI spec + example clients available in the developer portal.
CLI

Lightweight command-line for scripts, CI, demos, operator workflows, and audit automation.

huddle jurisdictions list
huddle compliance validate ...

Start in the Sandbox with validation, lifecycle reads, retrieval, HITL requests, drafts, and non-executing payment intents. Production credential issuance remains a reviewed access workflow.

How Agent + Human Collaboration Works

Transparent, auditable, with human control at every binding step.
  1. 1. Agent uses Huddle Skill → validates jurisdiction + terms and reads lifecycle state.
  2. 2.Agent creates HITL requests, bond drafts, or rent payment intents only within granted scopes.
  3. 3. Human or standing delegated authorization supplies consent before any binding or financial mutation.
  4. 4. Huddle verifies identity, payment validity, idempotency, chain state, and auditability before execution paths open.
  5. 5. Agent continues with lifecycle monitoring, compliance packet retrieval, and evidence workflows per granted scopes.
Platform operations included

Platform fees support Huddle's software, compliance operations, evidence, administration, and customer support. They are separate from bond principal.

Full Huddle Skill & Enterprise Agent API Reference (v0.2)

v0.2 Tool Catalog

Available (read-only / validation)
  • huddle.list_jurisdictions — Source-backed jurisdiction status
  • huddle.validate_deposit_terms — Legal + policy + control checks
  • huddle.get_lifecycle_status — Redacted Huddlebond lifecycle state
  • huddle.get_lease_payment_schedule — Lease-derived schedule projection
  • huddle.get_compliance_packet — Evidence lineage + hashes
  • huddle.get_commercial_bond_status — Workspace status summary
  • huddle.get_principal_protection_proof — Golden Rule / Sentinel proof
  • huddle.get_audit_events — Redacted workspace audit trail
  • huddle.get_erp_export — Yardi, AppFolio, QuickBooks, Xero, etc.
Available HITL draft creation
  • huddle.initiate_activation — Creates a durable draft + human handoff URL
  • huddle.accept_invitation_request — Requests human approval to accept an invitation
  • huddle.create_bond_draft — Creates a non-financial Huddlebond draft
  • huddle.request_bond_approval — Creates a HITL approval gate; no execution
  • huddle.deploy_bond — Approved-HITL landlord deployment on allowlisted Base Sepolia only
  • huddle.create_rent_payment_intent — Creates a non-executing rent payment intent
Still gated behind delegation + HITL + reconciliation
  • huddle.accept_invitation — Not callable in v0.2
  • huddle.fund_bond — Not callable in v0.2
  • huddle.record_rent_payment — Not callable in v0.2
  • huddle.setup_recurring_rent_authorization — Not callable in v0.2

Education: Humans + Agents

Agents using the Huddle Skill can dramatically reduce friction for their humans by validating terms, reading lifecycle state, and preparing safe requests. Humans retain control via KYC, explicit confirmation, or tightly scoped standing authorization. Every action is auditable and platform fees remain separate from bond principal.

Full "Huddle Skill Cookbook" and production portal coming in Phase 2.

Agent-Initiated Huddlebond Journey

How an authorized agent using the Huddle Skill prepares lifecycle work while Huddle keeps consent, identity, compliance, payment validity, and auditability in control.

Step 1: Discovery, Validation, and Lifecycle Read

Agent calls huddle.list_jurisdictions, huddle.validate_deposit_terms, and huddle.get_lifecycle_status to understand what is safe and allowed next.

Step 2: Safe Request, Draft, or Intent

The agent can prepare requests, drafts, and intents, then call huddle.deploy_bond only after approved HITL on allowlisted Base Sepolia. Mainnet, other chains, funding, charges, settlement, transfers, and ERP posting remain unavailable.

Step 3: Consent and Delegation Check

A human confirms the request, or Huddle verifies a standing delegated grant with the right principal, role, scope, spending limit, expiration, and revocation status.

Step 4: Reconciliation Before Execution

Before any future execution tool opens, Huddle reserves idempotency, prevents duplicate chain calls, validates payment state, and exposes reconciliation blockers through lifecycle status.

Key Guarantee: No financial mutation, chain write, settlement, ERP posting, or payment execution is exposed in v0.2. Future execution requires explicit delegated authority or fresh human approval, idempotency, reconciliation, and audit coverage.

The Human + Agent Collaboration Model

Huddle is designed for a world where autonomous agents act on behalf of humans — but never without them.

Agent Strengths
  • • Rapid compliance validation
  • • Pre-filling complex data
  • • Continuous monitoring & evidence retrieval
  • • 24/7 operational support
Human Control (Non-Negotiable)
  • • Identity verification (KYC)
  • • Final economic confirmation
  • • Scope & permission decisions
  • • All financial & legal commitments
Shared Outcome
  • • Full audit trail for everyone
  • • Golden Rule controls apply to financial actions
  • • Agents get powerful tools; humans stay in control
Phase 4 — Agent Ecosystem

MCP Federation & Skill Discovery

Agents don't operate in isolation. The Huddle platform enables secure, auditable discovery and composition between skills — while preserving the Golden Rule and explicit human oversight at every binding step.

Registry as Discovery Backbone

Every listed agent exposes machine-readable metadata (verticals, scopes supported, protection tier, composability hints). Future Huddle Skill versions can query the live registry at runtime to find peer agents for handoff or parallel validation.

Browse the live Agent Marketplace →
Agent-to-Agent Composition (with HITL)

Example flow: A PM Portfolio Orchestrator detects an exception in a commercial lease → invokes the specialized Commercial Bond Compliance Agent (or external Auditor Agent) for secondary review → both agents surface a unified evidence packet → human landlord or CFO gives final sign-off. Every hop is logged and contributes to Sentinel health.

All cross-agent calls respect redaction profiles and declared scopes.
Private Federation for White-Label Partners

Enterprise partners can run private MCP meshes and custom skills that still participate in the shared Huddle trust graph. Federation metadata lets internal agents discover both public Huddle skills and partner-private ones under a unified namespace.

Explore White-Label options →
Non-negotiable invariants (even in federated scenarios)
  • Human KYC + explicit final confirmation is always required for activation, fund movement, or legal commitments.
  • Every agent action (local or cross-skill) generates a signed, reproducible audit receipt.
  • platform membership allocation is collected at the platform boundary and funds the shared Sentinel — no agent bypasses this.
  • Redaction and scope enforcement remain server-side; federation only expands discovery, never trust boundaries.

Technical deep-dive and MCP federation spec available in the Enterprise Agent API docs and Huddle Skill reference.

Ready to Huddle Up?

Three fast paths for agent developers, enterprises, and humans whose agents are already working for them.

I'm an agent developer

Load the Huddle Skill, connect to the hosted MCP endpoint, and start validating + initiating in < 30 minutes.

I'm an enterprise / auditor

Get sandbox credentials for the Enterprise Agent API. Production tiers available with usage metering and SLAs.

I want my agent to onboard me

If your agent or landlord already started the process, use the agent-initiated path in the Activate gateway.

Huddle Agent Registry & Ecosystem

Official and partner-built agents that extend Huddle capabilities. All agents respect the same HITL guardrails and audit requirements.

Official — Huddle Skill v0.2

The canonical skill for lifecycle reads, HITL requests, drafts/intents, evidence retrieval, and approved-HITL testnet bond deployment. Available via MCP and package.

Get the Official Skill →
Example: Commercial Bond Compliance Agent (Partner)

Specialized agent for large commercial portfolios. Automatically validates complex lease-security terms, pulls ERP exports, and prepares annual compliance packets for auditors.

• Strong HITL on all financial actions
• Higher automation limits
• Portfolio-level monitoring
Explore the live Agent Marketplace (search + filters by vertical) or get started in the Developer Portal.
Living Platform

What's New & What's Next

Shipped in the last 4 weeks (Phase 4)
  • • Agent Registry → full marketplace with live search, vertical filters, Featured + Trending, richer cards with ratings/verified/composability signals
  • • 4-step self-serve Publish Wizard + live listing preview + partner-status dashboard (REG- IDs)
  • • Advanced commercial Volume & Protection Simulator in Developer Portal (tiered discounts, committed-use toggle, live Sentinel math)
  • • Live White-Label Configurator preview (real-time branding + domain)
  • • MCP Federation & Skill Discovery narrative + cross-agent composition model live across hub + registry
  • • Deeper discovery pills and clear "Huddle Up with this agent" handoffs

Draft initiation, pricing config, and handoff routes are Git-tracked implementation. Production build, tsc, and lint must pass before shipping.

Ecosystem Signals (planned)
When pilots are live and approved for public reference, this surface will publish aggregate outcomes (Sentinel contribution, exception rate, time-to-evidence). No partner data is shown today.
Want your pilot results featured? Reach out via the developer portal once production credentials are issued.
We will only publish metrics that are (a) measurable in our analytics pipeline and (b) cleared by the participating partner.
Next 6–8 weeks (public roadmap)
  • → CEO / legal review for evidence upload, approval, minting, ERP-posting, and fund-moving tools
  • → Federation metadata API + runtime skill discovery (Huddle Skill v0.2)
  • → Real sandbox credential issuance path (request → review → token)
  • → Live usage analytics surface for Sentinel allocation attribution
  • → White-label one-click launch (CNAME + private MCP) MVP

Priorities shift with partner and agent feedback. This is a living signal, not a contract.

This entire /for-agents surface (and the Activate gateway, Pricing, and dashboards) is the primary on-ramp for the 2026 flat-fee + Sentinel model. We optimize in public because agents and partners are first-class participants.
Built on the same Enterprise Agent API used internally by Huddle and pilot partners. MCP, REST, and CLI access with consistent authorization, redaction profiles, and audit logging. Platform fees support the software, compliance operations, and support behind these tools. Full documentation and examples are available for qualified users.