Demand-first AI agent marketplace

Build agents
that solve real
problems.

Most platforms start with supply — build an agent and hope someone wants it. AgentForge starts with demand. Someone posts what they need. Builders compete. Winning agents go to the marketplace where anyone can reuse them.

Demand → Build → Payment → Reuse
bounty.spec.ts 
// Post a bounty — describe what you need
const bounty = await forge.createBounty({
  title: "Lead enrichment agent",
  acceptanceCriteria: [
    "Takes company name as input",
    "Queries 3 data sources",
    "Returns structured JSON",
    "Handles 404 gracefully"
  ],
  prize: 1500,
  deadline: "2026-07-01"
})

// Submit your agent as a solution
const agent = await forge.submitAgent({
  systemPrompt: "You enrich company leads...",
  instructions: "Use web search first...",
  tools: ["web-search", "json-parse"]
})

// Win → agent goes to marketplace
await forge.publishToMarketplace(agent)
Live bounty feed
nlpresearch Document classifier agent $800 Open
automationslack Slack triage bot for support queues $1,200 In Review
codetesting PR review agent with CVE detection $2,500 Winner selected
dataapi Webhook → CSV transformer agent $650 Open
writingcms Notion page summarizer for content teams $900 Open
47
Active bounties
$94K
Paid to builders
312
Agents in marketplace
89%
Win rate for top builders
How it works

Four steps. Full loop.

01
Post a bounty
Describe the agent you need. Set acceptance criteria, prize amount, and deadline. No code required — just clear specs.
02
Builders compete
Developers submit agents with system prompts, instructions, tools, and repo links. Multiple builders, multiple approaches.
03
Pick the winner
Review submissions, test outputs, compare approaches. Simple scoring UI. Pay the winner directly.
04
Agent goes to marketplace
Winning agents become reusable assets. Set your price, earn on every use. Builders keep monetizing their work.
The marketplace

Every completed bounty becomes a reusable asset.

Agents in the marketplace come with proof of work — they solved a real bounty, passed real acceptance criteria. Not just a prompt dump.

⊕
Search by task type
Filter by tags, input format, output structure, tools required
⊕
Fork and customize
Clone any agent, modify for your use case. Builder reputation travels with it.
⊕
Monetize usage
Set per-call pricing. Earn every time someone uses your agent. Stripe-ready.
Lead Enricher
$0.05/call
Built by priya_dev · won $1,500 bounty · 89 calls this month
web-search json crm
78% win rate
PR Reviewer
$0.12/call
Built by marco_builds · won $2,500 bounty · 234 calls this month
security git cve
94% win rate
Notion Summarizer
$0.03/call
Built by elena_codes · won $900 bounty · 51 calls this month
notion writing cms
65% win rate
Agent builder

Form-based. No drag-and-drop.

Agents are defined by four things — system prompt, instructions, tools, and input variables. No visual nodes. No flowchart builders. Just code primitives that work.

The core identity and behavior of your agent
You are a lead enrichment agent. Given a company name, you query three public data sources, normalize the results, and return structured JSON with confidence scores.
Step-by-step guidance for handling edge cases
1. Query Clearbit for company data
2. Fall back to LinkedIn if Clearbit returns 404
3. Use Crunchbase only for funding data
4. Return null for any field you cannot fill
5. Always include confidence: 0.0–1.0 per field
APIs and capabilities your agent can call
web-search json-parser http-request + add
What the caller must provide
company_name string max_results number + add
Agents should earn their keep.
Not sit in a repo gathering dust after the project ends.
AgentForge exists to make AI agents useful — starting with someone who needs one, ending with an asset anyone can reuse.