Everything a production AI workflow needs
Composable by design
Build workflows as trees of nodes — agents, conditions, loops, code steps, and human steps — composable the same way code is.
Agentic where it counts
Give each step exactly as much autonomy as the task needs. Tool-call loops, retries, and LLM validation are built in.
Humans in the loop
Pause any workflow for human review, deliberation, or structured form input — text, numbers, dates, ratings, files, location, and more — then resume without losing state.
Voice-first workflows
Let participants complete surveys, approvals, and data collection by speaking — same branching, same validation, zero workflow changes.
Full observability
Every run produces a structured step log and scoped state tree. Know exactly what happened and why.
Any model, any tool
OpenAI, Anthropic, Gemini, and Ollama out of the box. Bring your own tools via HTTP endpoints.
Versioned deployments
Snapshot a workflow and its agents into a deployment. Ship changes safely without breaking live runs.
Works with what you already have
Add AI workflows to any application
Legacy systems, internal tools, customer portals — embed intelligent agents and structured workflows into the apps your team already runs. No rewrite required.
Serverless embed
One component, one prop. No backend, no token plumbing — just a deployment ID and your workflow is live on any page.
<ChatterflyWidget
deploymentId="dep_abc123"
/>Full widget embed
Mint tokens server-side, control run creation, pass custom input, and wire lifecycle callbacks — full control for authenticated connections.
<ChatterflyWidget
runId={runId}
participantToken={token}
/>Both paths give your existing app the same capabilities — AI agents, structured forms, human-in-the-loop review, and real-time chat — without touching your current codebase.
For developers & coding agents
Build workflows with code
Workflows are a JSON DSL — version them in git, validate them in CI, and let coding agents author them with full DSL context.
CLI
Validate, deploy, and trigger workflows from your terminal or CI pipeline.
chatterfly validate flow.json
chatterfly deploy flow.jsonCoding agents
An MCP server and llms.txt give agents the DSL reference, validation, and deploy tools out of the box.
{ "url": "$API/mcp" }REST API
A documented management API with an OpenAPI spec for workflows, runs, and connections.
POST /api/management
/workflows/validate