Ship AI agentspeople actually use.
AgentMason is a marketplace and studio for useful AI agents. Search by profession, install a real package, see the workflow, connect the tools, run evals, then share a chat link or API endpoint.
Search, build, and ship in three steps
Search the marketplace
Find a ready agent by job, role, app, or outcome in a curated library — or start from a blank intent.
Install or build from intent
Install a real package in a click, or describe one intent and let Agent Studio draft the workflow and auto-generate evals.
Use it as chat, API, or embed
Share a chat link, call the run API, or drop a chat-bubble widget on your site. Same agent, three surfaces.
One intent in. A working agent out.
Type a single plain-English intent. The studio drafts the visual orchestration — single, sequential, multi-agent, sub-agent handoff, retry loop, or human approval — auto-generates smoke evals, and suggests the right tools and connectors. The form is secondary; the picture comes first.
Start from a curated, deduped library
Browse ~2,000 dense, reusable agents by job, role, app, or outcome — not 47k of noise. Filter by category, sort by trust, and install in a click. Every listing resolves to a real package, not just metadata.
Every agent is files, not a mystery prompt
Each agent ships as a package: a visible workflow, runtime spec, README, smoke evals, an example run payload, and an env template. Inspect every file, run evals before publish, and export to Claude Code or OpenAI.
Use it three ways
The same published agent works as a shareable chat, a run API, or a chat-bubble widget you drop on any site.
Chat link
Share /chat/your-agent. Recipients chat instantly — no workspace, no login for public agents.
Run API & MCP
Call the same agent from code with a scoped API key, or expose it to any MCP-aware client.
curl -X POST \
"$BASE/api/agents/\
web-search-agent/runs?wait=true" \
-H "Authorization: Bearer $KEY" \
-H "Content-Type: application/json" \
-d '{"input":{"task":"Summarize AI news"}}'Embed widget
Drop one snippet on any site to get a floating chat-bubble widget in the corner.
<script src="$BASE/embed/your-agent">
</script>Everything you need to take an agent to production
Build, connect, evaluate, run, schedule, and share — all in one place.
Visual Agent Studio
Orchestration patterns from single agent to multi-agent handoff, loops, and human approval.
Curated library
~2,000 reusable agents across coding, research, finance, ops, design, data, and education.
Connectors
GitHub, Gmail, and Composio's 1,000+ toolkits via OAuth, plus any custom MCP server.
Built-in tools
Web search, browse, fetch, memory, calculator, sequential thinking, and code interpreter.
Evals
Smoke tests ship with every agent and run before publish so you know what works.
Runs & metrics
Trigger runs, watch the event timeline, and track success rate and latency.
Schedules
Run agents on a recurring schedule with notifications when they finish.
Drives
Upload files into searchable collections agents can read as grounded context.
Shareable chat
A focused public chat link recipients can use without ever seeing your workspace.
Embeddable widget
Drop a compact chat widget onto any website with a single snippet.
Run API + MCP
Call agents from code with scoped API keys, or expose them to MCP-aware clients.
Multi-model
OpenAI, Anthropic, Bedrock, Ollama, and OpenRouter — bring your own provider.
Which one are you?
Pick a ready agent, build your own, or operate them at scale.
Pick a ready agent and run it
Build from intent, inspect every file
Keep agents safe and observable
Build your first useful agent in minutes.
Start with search if you know the job. Start with the studio if you already know the intent.