Give your agent a job,
and the tools to finish it.

An agent is not a chatbot, it is a worker with tools. Tiny Agents runs on the same Studio canvas as your forms, tables and workflows, in purple. Give it instructions, pick how smart it should be, and hand it a set of tools. Then watch it think, pick the right tool, and pause for your approval before it acts.

Start building free Watch one run

Think, act, check, repeatReal tools, not guessworkSix agents ready to runAsks before it actsEmbed it on your site or in chatReads your tables and apps

tiny command · AgentsLead Qualifierdone

Plan the workthinking

“New lead from Acme Robotics. Research the company, score intent, update the CRM, then draft outreach, and get it approved before sending.”

Tiny Scouttool

result

{
  "company": "Acme Robotics",
  "intentScore": 82,
  "fit": "high"
}

Update Recordtool

result

{
  "table": "Leads",
  "status": "hot"
}

Human Approvalapproval

“Per the instructions, pause for a person to approve before anything is sent.”

A real run. Each step is a thought, a tool it picked up, or a pause for your approval.

Flagship

Drop an agent into your chat, or onto your site.

Take a finished agent and put it where your customers already are. Paste one line of code on your website, or add it to a chat. Give it your own knowledge as a PDF and it answers from your content, in your voice, and every answer points back to the page it came from. Each time you save, it keeps a version, so you can always roll back.

One line on your site

<script src="https://agent.tinycommand.com/widget.js"        data-agent="docs-assistant"        defer></script>

Answers from your PDFsReplies as you typeVoice questionsCites its sourceVersion history

Docs Assistantanswers from your knowledge

Hi, I am your docs assistant. Ask me anything about the product.

How do I connect a custom domain?

Open Settings, then Domains, and add a CNAME pointing to your space. It verifies in a few minutes.from setup-guide.pdf · p.4

Type a message↑

Watch it think, pick a tool, and check with you before it acts.

Given a goal, the agent does not answer in one shot. It makes a plan, picks a tool, reads what comes back, and decides the next move, repeating until the job is done. You can see the whole run: every thought, every tool it used and what it got, and a pause for your approval before it sends anything.

Lead Qualifier · the runasks before it acts

Plan the workthinking

“New lead from Acme Robotics. I will research the company first, score intent, update the CRM, then draft outreach, and get it approved before sending.”

output

{
  "plan": ["scout","update","compose","approve"]
}

Tiny Scouttool

given

{
  "company": "Acme Robotics",
  "industry": "Industrial automation"
}

result

{
  "summary": "Series B, hiring ops",
  "intentScore": 82,
  "fit": "high"
}

Read it and decidethinking

“intentScore 82 is at or above 70, so this is a hot lead. Mark it hot in the CRM, then write a tailored first-touch email.”

output

{
  "decision": "hot",
  "next": "update_record"
}

Update Recordtool

given

{
  "table": "Leads",
  "rowId": "rec_8841",
  "status": "hot"
}

result

{
  "ok": true,
  "rowId": "rec_8841"
}

Tiny Composertool

given

{
  "objective": "book a discovery call",
  "recipient": {name,title}
}

result

{
  "subject": "Scaling Acme ops",
  "body": "Hi Ada, saw Acme just"
}

Human Approvalapproval

“Drafted the email. Per my instructions, I pause for a person to approve before anything is sent.”

awaiting input

{
  "prompt": "Approve outreach to Ada?",
  "assignee": "sales-lead"
}

Approve & resume →Reject

Final answeranswer

“Approved. Lead qualified as hot, CRM updated, and a personalized first-touch email is ready to send.”

output

{
  "status": "qualified · hot",
  "rowId": "rec_8841",
  "email": "ready_to_send"
}

thinking · the agent planstool · it picks a tool and gets a resultapproval · it pauses for a personanswer · the finished result

Every action your platform can take is a tool the agent can pick up.

The agent is the brain. Tools are its hands. Web search, scraping, reading documents, looking up people and companies, your Tiny Tables, and your apps are all tools on the same canvas, the very same building blocks that power your workflows. Each one has a clear input and a clear output, so the agent runs a real action instead of guessing.

AllAgentBuildWebVisionDataActionHuman

Agent

Tiny Scout

Agent

Tiny Composer

Agent

Tiny AI

Agent

Tool Input

Build

Tool Output

Build

Formula

Build

If / Else

Build

Web Search

Web

Scrape Page

Web

Crawl Site

Web

Extract Data

Web

Map URLs

Web

Read Text

Vision

Read PDF

Vision

Pull Table

Vision

Pull Fields

Vision

Ask a Doc

Vision

Sort Docs

Vision

Add Record

Data

Update Record

Data

Look Up Data

Data

Call an API

Data

Person Lookup

Data

Company Lookup

Data

Email Finder

Data

Gmail

Action

Slack

Action

HubSpot

Action

Stripe

Action

Notion

Action

Any Integration

Action

Human Approval

Human

Give it company (a name)it runs the tool, andgets back summary and intentScore

Six agents ready to run today.

Skip the blank canvas. Drop in a research or outreach agent and it is useful on day one, or build your own with instructions and tools. Tiny Scout and Tiny Composer also live as first-class building blocks on the canvas.

Tiny Scout

Researches a prospect or a company and brings back a clear summary, an intent score, and a fit read.

Full Company ResearchOutbound ProspectingExecutive Briefing

Tiny Composer

Writes personalized messages and emails from a recipient and a goal.

OutreachFollow-upIntroduction

Tiny Search

Searches the live web and brings back grounded, sourced answers the agent can use.

Live web answer

Company Enricher

Fills out a company from a name or domain, drawing on multiple sources so the answer rarely comes back empty.

Company details

Person Enricher

Turns an email or a name into a full, organized profile.

Contact profile

Email Finder

Finds and checks a work email, switching between sources when one comes up short.

Find & verify

Set it up, give it tools, test it, then go live.

Every agent is built through the same three steps, and each one unlocks the next. You set it up first, then give it its tools, then test it on a real example and watch the run. You only go live with something you have seen work, not a blind guess.

①

Set it up

Name it and write its instructions, the personality and the job it does.

②

Give it tools

Choose how smart it should be and hand it the tools it is allowed to use. Turn on memory so it remembers what came before, across a chat and across runs.

③

Test it

Run it on a real example, step through the run, and confirm it does the right thing before customers ever see it.

↯

Go live

Ship the agent you just tested. The same engine that ran your test runs it for real.

tiny command · Agentson the canvasin a workflow

1 Set up2 Tools3 TestGo live →

Identity

Lead Qualifier

qualifies and routes inbound leads

Instructions

You are a sales development rep. For each new lead, research the company with Tiny Scout, decide if it is a fit (intent of 70 or more is hot), update the CRM record, then draft a first-touch email. Ask a person to approve before sending.

How smart

Tiny AI NanoTiny AI MiniTiny AI ProTiny AI Max

Memory remembers what came before, across a chat and across runs

Tools · 4

+ add tool

Tiny Scoutresearchresearch

give it a companyand an industrygets back a summary and an intent score

Update Recordwrites to your tableaction

in Leadsfind the rowset its status

Tiny Composerwrites the emailwrite

an objectiveand a recipientgets back a subject and body

Human Approvalpauses for a persongate

a promptand who to askpauses the run

The same canvas as forms, tables, workflows, and email. In purple.

Agents is the purple side of the one Studio canvas, and that is the whole point. An agent here can read your Tiny Tables and reach the same apps your workflows use, because it is the same engine underneath. A point tool bolted onto your stack cannot say that.

Formswarm orange

Workflowsdeep blue

this page

Agentspurple

Tablesgreen

Emailred

An answer bot, an app-action bot, or a worker.

Most tools are great at one half of the job. The difference here is a worker that thinks, picks the right tool, checks with you, and lives right next to your data, all in one place.

	Tiny Agents	Intercom Fin	Chatbase	Zapier Agents	Lindy
Thinks, acts, reads the result, repeats	✓	No	No	✓	✓
Tools with a clear input and output	✓	No	No	app steps	app steps
Pauses for your approval, then resumes	✓	handoff	No	limited	✓
Embed it on your site, answers from your PDFs	✓	✓	✓	No	limited
Same canvas as your forms, tables, workflows	✓	No	No	No	No
Reads your own data directly	✓	No	No	via apps	via apps
Web, document reading and lookups as tools	✓	No	scrape	via apps	some

If you want the best support deflection at scale, Intercom Fin is excellent. If you just need a knowledge-base chatbot on a page fast, Chatbase is great. If you need the widest app directory to hang an agent off, Zapier has it, and Lindy ships polished autonomous agents. If you want a worker that thinks, picks the right tool, checks with you, and lives right next to your forms, tables and workflows, that is what we built.

Good to know.

How is this different from a chatbot? +

A chatbot answers from a knowledge base. A Tiny Agent does work. It makes a plan, picks the right tool, reads each result, and decides the next step, repeating until the goal is met, with a pause for your approval before anything important happens. And you can step through the whole run to see exactly what it did.

What do you mean by "tools"? +

A tool is a real action the agent can take, like looking up a company, searching the web, or updating a row in your table. Each tool has a clear input (what you give it) and a clear output (what it gives back), so the agent runs a real action and gets a tidy result, instead of guessing.

How smart can it be? +

You choose how much brainpower the agent gets, from Tiny AI Nano for quick, simple jobs up to Tiny AI Max for the hardest reasoning. Pick the level that fits the task and your budget.

Can it pause and ask a person before acting? +

Yes. Human Approval is a real step you can add. Drop it in and the agent pauses for someone to approve, review the data, or fill something in before an important action, then picks up right where it left off once you respond.

Can the agent reach my data and apps? +

On the Studio canvas, the same building blocks that power your workflows become tools, so an agent can read and write your Tiny Tables, look up a person or company, search the web, read a document, and use your connected apps. When you embed the agent on your own site, you connect the specific actions you want it to take, so check the one you need before you ship.

Do I have to start from scratch? +

No. Six agents are ready to run, including Tiny Scout for research and Tiny Composer for outreach. Use one as is, or build your own by setting it up, giving it tools, and testing it.

Is there a free plan? +

Yes. Build and test an agent for free, and you are in the builder in under a minute.

How it all works together