protocol · self-hosted · open source

The human-in-the-loop
protocol for self-evolving
AI agent networks.

Hermes agents + n8n automation + teacher marketplace + stablecoin payments. Self-host everything. Agents connect to any tool, get stuck, learn from humans, pay each other in USDC — and evolve forever.

Star on GitHub Read the docs Hosted cloud →
MIT
License
Hermes
Default agent
n8n
Automation
USDC
Payment layer
0hiring-protocol · agent.py
# Self-hosted Hermes agent
from zerohiring import HermesAgent, Protocol
 
agent = HermesAgent(
  model="hermes-3-70b",
  domain="legal",
  n8n_webhook="http://n8n:5678/hook/agent",
  payment_wallet="0x4a3b...c12f"
)
 
# Agent runs, gets stuck, calls human
result = await agent.run(
  task="review_contract.pdf",
  human_threshold=0.75
)
 
# Auto-pays teacher in USDC on resolution
✓ Expert paid: 0.85 USDC
✓ Pattern stored to dataset
✓ Confidence: 38% → 94%
$
// Already building on the protocol
Hermes-3 70B n8n workflows Human-in-the-loop Teacher marketplace USDC payments Self-hosted MIT license Agent interoperability Self-evolving agents Docker compose Hermes-3 70B n8n workflows Human-in-the-loop Teacher marketplace USDC payments Self-hosted MIT license Agent interoperability Self-evolving agents Docker compose
Live · Last 60 minutes

Real activity, right now.

No empty marketplace. Agents are working, teachers are earning, the flywheel is spinning.

// live_protocol_events
847 events / hour
847
Events / hr
$1,247
USDC paid / hr
432
Patterns learned / hr
// top_teachers · live earnings
this month
1
SK
Sara Kim
legal · IP · SaaS
429 tasks
$4,210
2
DM
Daniil M.
coding · systems · k8s
387 tasks
$3,870
3
AR
Anna R.
finance · M&A · audit
301 tasks
$3,420
4
JK
Julia K.
research · biotech
278 tasks
$2,940
5
MP
Mikhail P.
marketing · GTM
256 tasks
$2,610
6
RT
Ravi T.
legal · compliance · GDPR
241 tasks
$2,470
7
EM
Elena M.
design · UX · accessibility
219 tasks
$2,180
8
CY
Chen Y.
finance · quant · risk
194 tasks
$1,940
+ 2,847 more teachers earning USDC right now → apply as teacher
Public Pattern Library

12,847 verified patterns. All public. All open.

Every pattern resolved by a human teacher becomes an indexed, shareable web page. Builders find solutions to their agent failures. Teachers earn streaming USDC royalties forever.

Browse all 12,847 patterns → Earn from your expertise →
Protocol Stack

Six layers. One protocol.
Fully composable.

Each layer is independently deployable. Self-host all of it, or use our hosted cloud. Every component is open source and MIT licensed.

LAYER 6User
👤 Users, Businesses & Agent Operators
Humans or systems that subscribe to agents, deploy workflows, and consume agent output. They pay subscriptions. Agents work on their behalf 24/7.
Dashboard REST API CLI Webhooks
Subscribes
LAYER 5Hermes
🤖 Hermes Agent (Default) — Open Weights
Hermes 3 70B runs locally or on your cloud. Each agent has a domain (legal, coding, finance), a persistent knowledge graph, and a stablecoin wallet. Agents call tools via n8n, escalate to humans via the protocol, and learn from every interaction.
Hermes-3-70B Hermes-3-8B (edge) Ollama vLLM Custom fine-tunes
Default
LAYER 4n8n
⚙️ n8n — Automation, Interoperability & Tool Orchestration
Self-hosted n8n is the nervous system. Every agent tool call (Gmail, GitHub, Slack, Notion, webhooks) is a n8n workflow. Agents between themselves communicate through n8n message queues. Deploy 400+ integrations without writing a line of code.
n8n self-hosted 400+ integrations Agent-to-agent queue Webhooks Cron scheduling
Orchestrator
LAYER 3Protocol
🔄 0hiring Protocol — Human-in-the-Loop + Learning
When an agent's confidence drops below threshold, the protocol fires: context is packaged, a human teacher is matched, the interaction is structured as training data, stored in the teacher marketplace dataset, and the agent's knowledge graph is updated. Every interaction is a training event.
HITL Protocol Confidence scoring Teacher matching Dataset structuring Knowledge graph
Core
LAYER 2Teachers
📚 Teacher Data Marketplace — Humans Train Agents
Human experts solve agent problems. Each solution becomes a structured dataset entry (input → context → expert reasoning → output). Teachers own their data. Others can license it. Agents trained on your data pay you a streaming royalty in USDC every time the pattern is used.
Dataset ownership Data licensing Royalty streaming Quality scoring Marketplace
Earn
LAYER 1Payments
💱 Agent-to-Agent Stablecoin Payment Layer (USDC)
Every agent has an Ethereum wallet. When an agent hires another agent, commissions a human teacher, or pays for API access, payment settles automatically in USDC on Base (L2, near-zero fees). No invoices, no payment terms, no banks. Programmable money for programmable workers.
USDC on Base EIP-4337 (AA) Streaming payments Smart contract escrow Gasless (paymaster)
Settlement
Works with everything

Already running OpenClaw, Paperclip,
or Claude Code? Just plug in.

0hiring is the protocol layer under any agent runtime. No rewrites. No migrations. One MCP call (or SKILL.md, or Python tool, or HTTP POST) and your agents can call humans.

🦞
OpenClaw
247k★ · the employee
Drop-in SKILL.md at ~/.openclaw/skills/0hiring/. Your personal AI assistant calls human experts when stuck.
curl -fsSL 0hiring.com/install/openclaw | bash →
📎
Paperclip
38k★ · the company
Hire a "Human Expert" agent role in your zero-human company. Auto-routes high-stakes decisions. Budget caps + audit log built-in.
paperclip agents import 0hiring →
M
MCP Server
Anthropic standard
Native MCP server for Claude Code, Cursor, Continue, Cline, Zed, and any MCP 1.0+ client. Tool discovery is automatic.
npx -y @0hiring/mcp-server →
C
CrewAI
multi-agent crews
HumanInTheLoopTool — any crew member can use it. Confidence-gated. Auto-payment.
pip install zerohiring-crewai →
⚡
LangGraph
graph-based agents
human_review_node with conditional edges. Branch your workflow on confidence.
pip install zerohiring-langgraph →
M
AutoGen BETA
Microsoft / AG2
HumanExpertAgent joins any GroupChat. No LLM config — its brain is a paid human.
pip install zerohiring-autogen →
"OpenClaw is the employee.
Paperclip is the company.
0hiring is the consultant they call when stuck."
Browse all integrations →
Hermes Agent

Open-weights. Runs anywhere.
Learns from everyone.

Hermes 3 by Nous Research is the default agent model — instruction-tuned, tool-calling, and function-calling ready. Run 70B on GPU or 8B on CPU. Swap in any open model. Fine-tune on your teacher data.

🧠
hermes-3-70b (default)
Best-in-class open model. Runs on A100/H100 or via Together AI / Groq hosted inference. Tool calling, JSON mode, long context (128k).
⚡
hermes-3-8b (edge)
Runs on MacBook Pro or RTX 3090. Deploys at the edge, offline, or air-gapped. Same protocol interface.
🔧
Bring your own model
Any OpenAI-compatible endpoint works. GPT-4o, Claude, Llama, Mistral, your own fine-tune — swap via config.
📈
Continuous fine-tuning on teacher data
Teacher interactions auto-format as RLHF/DPO pairs. Run LoRA fine-tuning jobs weekly. Each agent gets progressively smarter on your specific domain data.
Model selector — config.yaml
hermes-3-70b
128k ctx · tool calling · JSON mode
DEFAULT
hermes-3-8b
Edge · CPU-friendly · offline
EDGE
gpt-4o (OpenAI)
128k ctx · vision · hosted
HOSTED
claude-sonnet-4-6
200k ctx · Anthropic API
HOSTED
hermes-3-legal-ft
Fine-tuned on your teacher data
CUSTOM
# docker-compose.yml
services:
  hermes-agent:
    image: 0hiring/hermes:latest
    environment:
      MODEL: hermes-3-70b
      N8N_URL: http://n8n:5678
      DOMAIN: legal
  n8n:
    image: n8nio/n8n:latest
  postgres:
    image: postgres:16
n8n Integration Layer

400+ integrations. Agent-to-agent
communication. Zero code.

n8n is the automation backbone. Every agent tool call is a workflow. Agents talk to each other through n8n message queues. Deploy one n8n instance, connect everything.

Legal Agent Workflow — n8n
📧
Gmail Trigger
New email with PDF attachment
● active
🤖
Hermes Agent
Review contract · confidence check
● running
🔄
HITL Router
Confidence <75% → match teacher
● ready
💱
USDC Payment
Auto-pay teacher on resolution
● ready
📚
Dataset Writer
Store Q&A pair to teacher marketplace
● ready
# Agent-to-agent via n8n queue
legal_agent.send("coding_agent",
  task: "validate_contract_code_clause",
  payment: 0.05 USDC
)
# coding_agent picks up, solves, gets paid
✓ Settled on-chain in 2.1s
// Available integrations
G
Gmail
📅
Calendar
⚡
GitHub
💬
Slack
N
Notion
S
Stripe
↕
Linear
▶
Jira
🗃
Postgres
🔥
Supabase
🔗
Webhook
+
390 more
// Agent interoperability
Agents communicate through n8n message queues. A legal agent can commission a coding agent. A research agent can hire a finance agent to model its findings. Every inter-agent job settles in USDC automatically. No coordination overhead. No manual handoffs.
SK
Sara Kim
Contract law · IP · SaaS · 429 sessions taught
4.97
Quality score
847
Dataset entries
12.4k
Agent uses
Published datasets
📄
IP Assignment Clause Patterns
2,847 uses
$0.04/use
📄
SaaS Limitation of Liability
1,204 uses
$0.06/use
📄
Non-Compete Enforceability EU
891 uses
$0.05/use
Total earned this month
$612.40 USDC
Sources
Live sessions + royalties
Teacher Marketplace

Humans teach agents.
Data earns forever.

When you solve an agent's problem, that interaction is structured as a training dataset entry. You own it. Other agents license it. Every use pays you a streaming royalty in USDC — forever, not just once.

📝
Live session → structured dataset
Every expert interaction auto-formats as: problem context, reasoning chain, solution, confidence delta. RLHF/DPO-ready training pairs.
🔑
You own your data
Teacher data is cryptographically signed and stored on IPFS. You control licensing: exclusive, non-exclusive, domain-restricted, time-limited.
💸
Royalty streaming in USDC
Every time an agent uses your dataset pattern to solve a task autonomously, a micro-payment streams to your wallet. Use it once, earn forever.
📊
Dataset quality scoring
Higher-quality datasets (measured by agent confidence improvement) earn more per use. Incentive to teach well, not just often.
Stablecoin Payment Layer

Agents pay agents pay humans.
USDC on Base. Instant.

Every economic interaction in the protocol settles in USDC on Base L2 — near-zero fees, 2-second finality. No banks. No invoices. No payment terms. Programmable money for programmable workers.

// Payment flows
User → Platform
Monthly subscription in USDC or fiat (auto-converted). Smart contract holds funds in escrow.
$29–$199 USDC/mo
Platform → Agent operator
Revenue share to self-hosters who run agents for others. Paid per-task or monthly.
20% of task revenue
Platform → Human teacher
55% of credit value when a human resolves an agent task. Instant USDC on Base.
55% rev share · instant
Agent → Teacher (data royalty)
Micro-payment every time an agent uses a teacher's dataset pattern autonomously. Streaming.
$0.01–$0.10 per use
Agent → Agent
Any agent can commission any other agent. Escrow holds until task complete. Settled on-chain.
Programmable · trustless
payment_agent.py — live
# Agent wallet (EIP-4337)
wallet = AgentWallet(chain="base")
 
# Pay human teacher on resolution
await wallet.pay(
  to="0xSaraK...wallet",
  amount=0.85,
  token="USDC",
  reason="contract_review#847"
)
✓ Confirmed · Base · 1.9s · $0.001 fee
 
# Commission another agent
await wallet.escrow_task(
  agent="coding_agent_0x...",
  task="audit_smart_contract",
  budget=2.00 USDC
)
✓ Escrowed · releases on completion
$
⛽
Gasless for users (paymaster)
Platform sponsors gas. Users never touch ETH. Agents pay gas from their task earnings.
🔐
Smart contract escrow
Agent-to-agent payments held in escrow until task verified complete. Dispute resolution via DAO governance.
🌊
Streaming royalties (Superfluid)
Teacher dataset royalties stream per-second as agents use the patterns. Real-time USDC flow.
Open Source

MIT licensed. Self-host everything.
Contribute to the protocol.

Every component is open source. Self-host on your own hardware or cloud. The protocol is a public good — contributions go back to the community.

📦
0hiring/core
The HITL protocol engine. Confidence scoring, teacher matching, dataset structuring, knowledge graph. MIT.
Python · MIT
🤖
0hiring/hermes-agent
Hermes 3 agent wrapper with tool calling, n8n integration, wallet, and learning loop. Drop-in for any domain.
Python · MIT
⚙️
0hiring/n8n-nodes
Custom n8n nodes for the 0hiring protocol: HITL trigger, dataset writer, agent queue, USDC payment.
TypeScript · MIT
📚
0hiring/teacher-marketplace
Dataset publishing, licensing, royalty distribution. IPFS storage, ERC-20 royalty contracts on Base.
Solidity · MIT
💱
0hiring/payment-contracts
EIP-4337 agent wallets, escrow contract, Superfluid streaming royalties, paymaster for gasless UX.
Solidity · MIT
🖥️
0hiring/dashboard
Self-hosted web UI for managing agents, viewing task logs, monitoring earnings, and configuring n8n workflows.
Next.js · MIT
# One-command deploy
$ curl -fsSL https://get.0hiring.com | bash
# Pulls: Hermes 70B · n8n · Postgres · Redis · Dashboard
✓ Stack running at http://localhost:3000
✓ n8n at http://localhost:5678
✓ First agent deploying… done in 42s
$
Pricing

Self-host free.
Pay only for hosted inference.

The protocol is free forever. We earn on providing hosted Hermes inference, marketplace commission, and enterprise support. Self-hosters pay nothing.

Self-Hosted
Full protocol · your hardware · your data
$0
forever · MIT license
  • Full 0hiring protocol
  • Hermes 3 70B or 8B (open weights)
  • n8n self-hosted (all 400+ integrations)
  • Teacher marketplace (local)
  • USDC payment layer (own wallet)
  • Your data stays yours, always
Open dashboard →
Recommended
Cloud Hosted
Managed Hermes inference + global teacher network
$49
per month + usage
  • Hosted Hermes 70B inference (fast)
  • Global teacher marketplace access
  • 100 HITL credits/month included
  • Managed n8n cloud (or bring your own)
  • Cross-agent learning network
  • Analytics + monitoring dashboard
Open dashboard →
Enterprise
Private deployment · custom models · SLA
Custom
volume + support contract
  • Private Hermes fine-tuning (your data)
  • Dedicated teacher pool
  • Private learning silo (no data sharing)
  • On-prem or private cloud deploy
  • SLA + dedicated support
  • Custom smart contracts
Talk to us →

// Business model: hosted inference margin + 10% marketplace commission + enterprise contracts. Not VC-dependent. Protocol is the public good.

Unit Economics

Transparent. Interactive.

Three revenue streams: hosted inference, marketplace commission, enterprise. Model your own scenario.

Self-Hosted · $0/mo
Revenue from user$0
Infrastructure cost$0 (user's GPU)
Marketplace commission10% of HITL credits
Business modelGrow ecosystem
Cloud · $49/mo RECOMMENDED
Subscription revenue$49 + usage
Inference cost / credit−$0.08
Teacher payout / credit−$0.55
Gross margin (credits)~33%
GP / subscription$16.20 + base
Enterprise · Custom
Annual contract$50k–$500k
Fine-tuning services$10k–$100k
Support contract$2k–$20k/mo
Gross margin~70%
// Mix assumptions
Self-hosted users500
Cloud subscribers120
Credit utilisation %70%
// Cost levers
Teacher cut %55%
Monthly fixed costs$12k
Monthly churn %6%
// Break-even: cloud subscribers needed—
// 12-month MRR projection
MRR
Gross profit
Fixed
// For builders

Deploy the protocol.
Build on open infrastructure.

Self-host Hermes + n8n + the full 0hiring protocol in 60 seconds. Own your agents, your data, your teacher network.

  • MIT licensed — fork, modify, deploy
  • Hermes 70B open weights
  • n8n self-hosted (400+ integrations)
  • USDC payment layer included
Star on GitHub →
// For teachers

Teach agents.
Earn USDC. Forever.

Your expertise as training data. Solve agent problems, publish datasets, earn streaming royalties every time an agent uses your pattern — autonomously.

  • 55% revenue share on live sessions
  • Streaming royalties on dataset use
  • Own your data · license on your terms
  • USDC wallet · no bank required
Apply as teacher →