FAQ

Is SMELT a framework?

No. SMELT Studio is a hosted workflow platform: drag-and-drop multi-agent workflows with managed AI, real-time execution monitoring, and a growing set of OAuth integrations. No infrastructure to manage.

Why not LangGraph / CrewAI?

Those tools optimise for demos and fast iteration. At production scale, teams often fight the abstraction. SMELT keeps orchestration in plain asyncio with a Pydantic state machine you can read, test, and modify. Sub-agent delegation is a canvas capability, not a framework API: no graph code to write and nothing to migrate away from.

How is SMELT Studio different from n8n?

n8n is excellent for generic automation - Slack messages, spreadsheet syncs, webhook chains - and it reaches far more applications than we do. SMELT Studio is built for AI workflows that need reasoning, planning, and quality validation, from a palette of purpose-built AI blocks: planning, saved agents, research, reasoning, retrieval, a direct model call, external delegation and content safety. An agent delegates to your other agents inside a single run, and the Trace shows every hand-off with what it cost in credits. Each workflow block pins an exact published agent version, so publishing a new version leaves existing workflows alone until you promote them. Different tools for different problems; the full comparison covers where n8n is the better buy.

How is SMELT Studio different from Dify?

Dify is a solid visual AI builder. SMELT Studio differs in three ways: purpose-built AI blocks rather than only generic LLM nodes, fully managed infrastructure (no Docker or self-hosting required), and flat monthly pricing rather than per-token charges at the platform level. Both are visual builders - SMELT is more opinionated about agent architecture. Delegation and retrieval are typed and visible in every run, and agent memory is yours to inspect and clear, not sealed behind the platform.

How do I get started?

Sign up for a free account at SMELT Studio. The free tier includes 3 workflows and 100 executions per month - no credit card required.

What is SMELT Studio?

SMELT Studio is a hosted web platform for building and running multi-agent AI workflows. Design workflows visually with a drag-and-drop canvas, connect to major providers and HTTP/MCP endpoints, and monitor execution in real-time. No infrastructure to manage - we handle the AI models, message bus, and compute.

Why NATS instead of Kafka?

Operational simplicity and developer velocity at early scale. JetStream still gives durable delivery and replay. If you outgrow it, the spec documents a migration mindset - SMELT does not bake in irreplaceable NATS semantics in your business logic.

Why three memory systems?

Redis, Postgres, and pgvector solve different problems: ephemeral working sets, durable run history, and semantic similarity. Collapsing them creates hot spots and expensive queries. SMELT names the split so your team does not pay that tax later. Agents can also keep their own memories across runs: you switch it on per agent, and you can see and delete everything an agent remembers.

What do I get for free vs paid?

The Free tier includes 3 workflows, 100 executions per month, and 75,000 managed AI credits on standard models. Pro (£19.99/mo) gives you 15 workflows, 10,000 executions, and 1,000,000 managed AI credits. Business (£59.99/mo) scales to 50 workflows and 50,000 executions. See the pricing section for the full comparison.

How does purchasing work?

Sign up free, then upgrade to Pro or Business via Stripe Checkout from your Studio dashboard. All subscriptions are monthly with no long-term commitment.

What happens if I cancel?

Your workflows stay where they are. You can downgrade to the Free tier at any time and keep the free tier workflow allowance. Run history is kept for 30 days from the run, on every tier, and cancelling does not change that.

Where is support?

Free tier users get the community Discord (#templates, #support). Pro and Business subscribers also get email support at [email protected]; Business subscribers receive priority response.

What are workflow templates?

Templates are pre-built workflows for common use cases: email classification, scheduled digests and research briefings, customer support triage, summarising and triaging over HTTP, and calling a tool on an MCP server. Each is a complete, working workflow with its blocks, prompts and connections already set. Run one as it comes or change it to suit your work - they are starting points, not constraints. See the Studio Templates page for the full list, and what each one needs before it runs.

What is the difference between Pro and Business?

Pro (£19.99/mo) is for individual builders: 15 workflows, 10,000 executions, and 1,000,000 managed AI credits. Business (£59.99/mo) is for teams: 50 workflows, 50,000 executions, 4,000,000 managed AI credits, plus shared team workspaces and priority support. See the pricing section for the full comparison.

Can I trigger a workflow from my own code?

Yes. Give a workflow a Webhook trigger and Studio generates a unique ingest URL with its own signing secret. Send a POST request to that URL with the secret in the x-webhook-secret header to start a run from a CI/CD pipeline, a script, or another service. The URL and secret live in the Webhook trigger's settings, with a copy button and an example.

If your sender retries, set an Idempotency-Key header to a value unique to each delivery. A repeat of a key we have already seen is answered 200 and starts no second run, so a retry costs nothing. Keys are remembered for 24 hours. Without the header two identical bodies are treated as two genuine deliveries and both run.

Still stuck? Start from Getting Started and trace one task through Architecture.