Skip to main content
FlowFn
Sign inStart free
AI Agents

AI agents that finish a task end-to-end.

Give the agent a task, pick the tools it can use, decide where it lives — a public chat widget on your site, a webhook your customers post to, or a cron that runs nightly. The agent iterates AI → tool → AI → tool until it produces a final answer.

Public chat widgetBYOK AIPlan-gated tools
Start freeHow it works
See it run

Visitors chat. The agent reaches into your stack.

Drop the chat widget on your site. A visitor asks a question; the agent reads your support files, queries your tools, hands off to a human when it needs to, and answers — keeping the conversation in the visitor's browser.

Core capabilities

Pick the surface. Pick the tools. Go.

Four agent types, one runtime.

Interactive (chat)

Public URL + embeddable chat widget for your website. Multi-turn conversation history, visitor identification optional, branding fully configurable.

Webhook

POST to a per-agent URL from your platform or a customer's. Signature-verified (per-platform or HMAC), token-rotatable, replayable from the run page.

Scheduled

Cron-driven agents for periodic work — daily summaries, hourly checks, weekly reports. Same runtime as interactive; the trigger is just a clock.

Playground-bound

When a playground end-user submits inputs, the bound agent runs. Useful for in-app AI features your customers interact with directly.

Human approvals

Approval rules can pause any agent run mid-iteration. The team gets the conversation context + the proposed next step, approves or rejects, the run resumes — or stops cleanly.

Support files + calibration

Drop in PDFs, docs, web pages or JSON. Calibrate the agent against your task once — it summarises what matters, you approve, and every run starts grounded in your business context, not the model's training data.

Runtime loop

AI picks a tool. The tool runs. AI picks the next one.

The agent runtime is a loop: the model reads the task + the conversation so far, decides what to do next — answer, call a tool, ask a human — then the runtime dispatches and feeds the result back into the next iteration. It stops when the model emits a final_answer.

Every step is persisted: the model's thought, the tool it picked, the input it sent, the output it got, latency, tokens used. You can inspect any run as a timeline or as a conversation.

Interactive agents have an extra guardrail: high-impact actions (delete, refund, mass-broadcast, role assignments, e-signature send) and entire high-impact platforms (Stripe, AWS, accounting, HR, raw databases) are filtered out of the picker — even if the team has them connected. Webhook, schedule, and playground agents keep full access.

Pair with workflows

Agents trigger workflows. Workflows can fire agents.

Agents and workflows are designed to compose. An agent can call any workflow you allow-list — useful when you want the agent to decidewhether to run something, and the workflow to handle how.

A workflow can fire an agent the same way, passing structured input. Both surfaces share the same run engine, BullMQ queue, audit trail, and BYOK accounting.

Trigger

New form submission

Contact form · 12 today

AI step

Classify intent

gpt-4o-mini · returns category

Tool action

Slack · Post to #sales

Routes pricing leads to sales

Run completed · 1.4s
Real teams use it for

Conversations and one-shots, on one runtime.

Customer-facing support

Embed the chat widget on your site. The agent reads your docs as support files, answers L1 questions, and hands off to a human via approval when it can't resolve.

Embed widget

Inbound webhook triage

A platform posts an event (Stripe, GitHub, Slack). The agent decides what matters, calls the right tool, and writes a summary back to your system — signature-verified end to end.

Webhook agents

Nightly digest agent

A scheduled agent pulls yesterday's data, asks itself what changed, drafts a digest, and posts it to the team Slack — pausing for human review when a number looks off.

Cron + approval

In-app AI feature

A playground-bound agent powers an end-user action inside your dashboard — generate, summarise, draft. The end-user never sees an LLM token; your team pays the bill.

BYOK AI

Plan-gated, not feature-gated.

Free tier ships with 1 agent + interactive surface. Pro and Business unlock more agents, webhook + schedule + playground-bound types, higher run-per-day caps, white-label embed branding, and content moderation appeals.

See pricing →

Ship your first agent today.

Pick a template, plug in the tools the agent can use, drop the widget on your site. No setup fee. No fine print. You can be live in an afternoon.

Start freeRead the docs
Contact

We’re here to help you ship

Email sales for volume pricing and enterprise procurement, reach support for everything else, or drop us a message and we’ll get back to you.

Talk to sales

Volume pricing, security review, and enterprise procurement — email us and we’ll get back fast.

General inquiries

support@flowfn.com

Prefer not to open your mail client?