AgentHub MVP — public beta

Trusted AI agents that actually ship work.

A marketplace for human-approved, evidence-cited, audit-ready agents that plug into your existing systems. No black-box autonomy. Every external write goes through hash-pinned approval.

Browse marketplace Open dashboard

Hash-pinned approvals on every external write
Source-cited answers with span-level evidence
Tenant-isolated retrieval (RLS + Qdrant filters)
Vercel AI Gateway — no direct provider SDK calls

Vercel AI Gateway

All model calls

Better Auth

Sessions + RLS

Neon Postgres

Tenant-isolated

Qdrant Cloud

Hosted vector RAG

Infisical

Secrets source-of-truth

CyberSource

Payments

Trust before breadth

Every capability designed so workspaces, admins, and compliance reviewers can trace, approve, and audit what agents do.

Hash-pinned approvals

Every external write, send, submit, delete, purchase, or admin action is canonicalized, SHA-256 pinned, and re-verified at execution. Mutated payloads are refused.

Cited retrieval, not vibes

Hosted Qdrant Cloud with mandatory tenant payload filters. Answers carry chunk_id, page, source span — never an array index.

Vercel-native runtime

Next.js, Fluid Compute, Workflows for durable runs, AI SDK 6 agents, AI Gateway for unified model access, budgets, and fallbacks.

Infisical-backed secrets

Provider OAuth client secrets, gateway keys, payment credentials, and webhook signing secrets live in Infisical. Vercel envs only consume references.

Pipedream Connect kernel

Embedded OAuth/key-based connections, MCP tool exposure behind AgentHub policy, audit, and approval gates. Slack, GitHub, Microsoft, Google.

Eval-gated releases

Golden datasets, citation verifier, prompt-injection corpus, tenant-isolation canaries. No agent ships without passing the quality gate.

Approval inbox

See exactly what the agent will do — before it does it.

Every risky tool call gets canonicalized, hashed, and queued for human approval. Diff view, recipient, target system, model trace, and risk reason — all on one screen.

View approval inbox Browse agents

approval_req_8f3c…2a91

risk: high

Tool

slack.chat.postMessage

Recipient

#cs-escalations · acme-prod workspace

{
  "channel": "C0L8N9R7P",
  "text": "Customer 4827 escalated — refund of $1,240 approved",
  "thread_ts": "1714425088.001"
}

sha256:9c1a…b740

Pricing

Pay only for what your agents ship. CyberSource-backed billing, internal entitlements, no Stripe.

Starter

Free during beta

1 workspace, up to 3 seats
Document Q&A pack
100 runs / month
Approval + audit baseline

Get started

Ship agents your compliance team can sign off on.

Start with the Document Q&A pack, install in under five minutes, and turn on more packs as your trust grows.

Browse marketplace Open dashboard