Magic Agent · Edge runtime
Production agents. At the edge.
Magic Agent is the agent runtime your product talks to. Sub-50ms cold start. Persistent memory. Tools as TypeScript, callable from any AI client. From $19/mo.
Pricing
Three tiers. Pick your throughput.
Every tier ships with the same edge runtime, the same persistent agent, and the same tool surface. You're paying for messages, memory depth, and tool budget.
Starter
One agent on the edge. For solo founders, side projects, and proofs of concept.
- ✓ 1 deployed Magic Agent
- ✓ Conversation memory included
- ✓ Basic tool registry
- ✓ Fast open-source models + Gemini
- ✓ Custom subdomain
- ✓ Community support
Pro
Most popularFive agents with persistent memory + retrieval. For teams shipping real product.
- ✓ 5 deployed Magic Agents
- ✓ Database + semantic memory
- ✓ Custom tools + agent-to-agent calling
- ✓ Plugin marketplace access
- ✓ WorkOS SSO
- ✓ Email support
Scale
Unlimited messages, every storage tier. For production agents under real load.
- ✓ Unlimited Magic Agents
- ✓ Database + semantic memory + media storage
- ✓ Priority tool execution
- ✓ Higher rate limits
- ✓ WorkOS SSO + audit logs
- ✓ Priority support + Slack
How it works
Three steps from prompt to production.
Describe the agent
Tell us what the agent should know, do, and decide. We spin up a persistent agent on the global edge in seconds.
Wire your tools
Add tools as plain TypeScript or connect any tool integration. Stripe, Linear, Notion, your own APIs – all available as agent capabilities.
Ship it everywhere
Embed in your site, open it up to other agents, or call it from anywhere. Same agent, same memory, every surface.
Use cases
What people ship on Magic Agent.
Customer support agent
Grounded in your docs + ticket history. Hands off to humans gracefully. Logs every conversation to your CRM.
Sales SDR agent
Qualifies leads, schedules calls, drafts follow-ups. Plugs into your CRM and calendar with built-in tool integrations.
Internal ops agent
Runs on your private docs + Linear + Notion. Triages issues, drafts updates, keeps the team in sync.
Site concierge
Lives on your Magic Site. Helps visitors navigate, answers product questions, books demos, takes payments.
Comparison
Magic Agent vs the alternatives.
| Feature | Magic Agent | Tidio AI | Replit Agent | LangGraph |
|---|---|---|---|---|
| Runtime | Edge agent | Centralized SaaS | Container (always-on) | Self-hosted Python |
| Cold start | Sub-50ms | n/a | Seconds | Depends on host |
| Memory | Database + semantic memory | Conversation log only | App-defined | Bring your own |
| Tool model | TypeScript + tool integrations | Closed integrations | Code-first | Python tools |
| Pricing entry | $19/mo | $29/mo (Tidio Plus) | $25/mo (Core) | Free OSS, hosted +$ |
Bundle
Magic Agent ships inside every Magic Sites plan.
Spark includes a Magic Agent Starter at no extra cost. Studio includes Magic Agent Pro plus 50% off Magic Workshop Solo for your first month. Agency includes three Magic Agent Pro seats plus a Magic Workshop Pro environment.
FAQ
Questions we hear a lot.
What's a Magic Agent vs a chatbot?
A chatbot is a script that answers messages. A Magic Agent is a persistent edge agent with memory, tools, and a long-running identity. It can take actions, hold state, and call other agents – not just respond.
When should I pick Magic Agent (Edge) over Magic Workshop (Container)?
Pick Magic Agent for production deployments with predictable workloads – chat surfaces, tool callers, agent-to-agent endpoints. Pick Magic Workshop for development environments, long-running tasks, or workloads that need full Linux + your own runtime. See /agents for the decision flow.
Which models can I use?
Fast open-source models (Llama, Mistral, Gemma) and Gemini out of the box. We do not ship OpenAI or Anthropic keys by default – you can bring your own if your plan needs them.
How does memory work?
Starter ships with conversation history. Pro and Scale add a database for structured state and semantic memory for retrieval. Scale also gets media storage for files + transcripts. All storage stays in your account namespace.
Can I bring my own tools?
Yes. Pro and Scale support custom TypeScript tools and any tool integration (any AI client can talk to your agent). Starter ships with a curated basic registry – upgrade to add custom tools.
What about rate limits?
Starter is metered to 5,000 messages/month. Pro is 50,000. Scale is unlimited within fair-use. Burst limits and queueing apply across all tiers.
Ship a Magic Agent today.
Production-ready agent runtime, low-latency, with persistent memory and agent-to-agent calling out of the box.