# ClearMakes Agents

AI agents that know your business and work in your tools.
Build AI agents from your documents and your apps. Connect Gmail, Slack, HubSpot and any MCP server, answer your team's questions, automate the busywork, and put an agent on your website in one line.

A ClearMakes product. Full documentation: https://agents.clearmakes.com/docs
Support: support@clearmakes.com

## What it is

An "agent" is an AI assistant you build for one job — answering customers,
helping a team, handling a workflow. It answers from three things you give it:
knowledge (your documents and website), connected apps (Gmail, Slack, HubSpot,
any MCP server), and behavior (how you tell it to act). It answers from your
content; answers can still be wrong. The chat does not show which document
an answer came from.

Chatbots are called "agents" throughout the product.

## The app

Signed in, a bar across the top has four places:
- Home: this month's numbers.
- Agents: your agents, and where you create new ones.
- Chats: everything people have asked, across all agents.
- Starters: ready-made agents to start from.

Open an agent and its tabs appear under its name: Summary, Instructions,
Sources, Actions, Style, Go live, Chats, Settings, plus a Try it button.

## Building an agent

Create: pick a starting point (blank, personal assistant, business agent by
industry and use case, or a template). Give it a name and a main goal.

Instructions: the main goal (one or two sentences on what it is for; required), a
tone (a preset or your own wording), and free-text custom instructions. Safety
and grounding rules are always added and not shown.

Sources: upload documents of any kind (or a whole folder), and add your
website (it crawls up to 200 pages, 50 by default, pulling out FAQs, tables and
docs). Files show their indexing progress and can be retried, downloaded or
removed. How hard the agent searches — and how well it reads long or
interlinked documents — is tuned in Settings.

Company knowledge (Company Brain): a shared Google Drive folder indexed for the
whole workspace and kept in step automatically. Each agent sets a sensitivity
ceiling (Anyone / Staff / Leadership) and which folders it may read. Drive
documents are read-only in the app and managed in Drive. Access is applied
before anything is searched, so an agent never answers from a document above
its ceiling.

Actions: link apps (Gmail, Slack, Jira, HubSpot and many more) per agent —
a connection belongs to one agent. Connecting grants nothing until you open the
app and say what the agent may do. You can also add a custom tool: a web service
(REST API) or an MCP server (paste its URL); the app reads the actions it offers
and you choose which the agent may use. Keys entered are encrypted and never
shown again. In a conversation the agent decides when to use a tool, running it
alongside searching knowledge.

Style: display name, avatar, welcome message, up to three suggested
questions (auto-generated if left blank), an accent colour (from a palette, a
custom colour, or pulled from your website), and whether to show the
"Powered by ClearMakes" badge.

Settings: Active/Paused; AI model (Automatic, or choose manually — providers
include OpenAI, Gemini, and self-hosted models, plus your own key); how much
knowledge to use (Low/Recommended/High) with accuracy, large-document and
relationship options; feature switches (file upload, image input, charts,
connected apps and MCP, data analyst, personas, document generation); and what the agent records (feedback, visitor location).

## Going live

Publish gives four ways to put an agent in front of people, all reading the
agent's live settings so edits carry through:
- Website widget: a one-line script pasted before </body>, adding a chat bubble.
- Share link: a direct full-page chat link.
- Embed: an iframe.
- API: talk to the agent from your own code (see below).
A paused agent does not answer anywhere until switched back on.

How answers are made: each message is understood (quick question, knowledge
search, a tool action, or data analysis), knowledge and tools run together, and
the answer comes from whichever has the information. Charts
and spreadsheet analysis appear when the question calls for them and the
feature is on. Try it runs this exact path.

Use from Claude or ChatGPT: create an API key on an agent's Knowledge tab,
scoped to exactly what that agent can read. You get a personal connector link to
paste into Claude or ChatGPT — reading tools only (search knowledge, fetch a
document, list areas). The key can also be a bearer token for the REST knowledge
endpoints. The secret is shown once; keys can be renamed, expired, rotated or
deleted.

## Developer API

Two public endpoints, both streaming server-sent events. No API key: an agent's
id is its capability, like a visitor to your site; pause the agent to stop it.

Start a conversation:
  POST /api/v1/chats/with-message
  body: { "chatbot_id": "AGENT_ID", "content": "message" }
  The first event carries chat_id.

Continue:
  POST /api/v1/chats/{chat_id}/messages/stream
  body: { "content": "next message" }

Events are lines of "data: {json}" with a type. The important ones are token
(a piece of the answer, appended in order) and done (finished). Others —
chat_created, sources, status, tool_status, tool_calls, intent, chart,
generated_image, error — can be shown for progress or ignored.

Agent appearance:
  GET /api/v1/chatbots/{agent_id}/widget-config
  Public presentation only; never instructions, knowledge or credentials.

Errors: before the stream, JSON with a detail; during, an error event with no
done. 403 = agent paused, 404 = no such agent/conversation, 422 = missing
content, 500 = the turn failed (retry is safe).

Widget: paste from the Go live page —
  <script src="https://agents.clearmakes.com/agents.js" data-chatbot-id="AGENT_ID" defer></script>

Creating agents, uploading knowledge and managing apps need a signed-in session
and are done in the dashboard, not through this public API.

## Plans

One price worldwide; two months free on yearly billing.
- Free ($0): 1 agent, 30 conversations/month, 5 documents (up to 10 MB) per agent.
- Starter ($12/mo): 2 agents, 300 conversations, 25 documents (up to 25 MB), visitors can share files.
- Pro ($39/mo): 5 agents, 1,000 conversations, 3 members, 100 documents (up to 100 MB), connect your apps, no badge.
- Business ($99/mo): 15 agents, 2,500 conversations, 10 members, 300 documents (up to 250 MB), plus "Precise" answers.
- Enterprise: custom limits; write to sales@clearmakes.com.

"Connect your apps" covers connected apps, custom REST/MCP tools, Google Drive
sync, and developer API keys. "Precise answers" unlocks
higher-quality models, your own AI key, and self-hosted models.

A conversation is counted once when it starts; a precise conversation counts as
five. One conversation can have up to 20 AI replies on Free and 50 on paid
plans. You are warned at 80%; agents answer a
little past the limit, then new conversations wait until the next month or an
upgrade. Visitors only ever see "this assistant is currently unavailable".
Downgrading never deletes anything — extra agents are paused and knowledge keeps
answering; you just cannot create or upload more until back under the limit.

## Team & data

Workspace settings hold the team and plan. Invite people with a single-use link
that lasts seven days (you send it yourself), as an Admin or a Member. Roles
also cap how much shared company knowledge someone's agents can reach.

Documents and websites are read to answer and never shared between workspaces.
Connected apps are used only by their agent, on behalf of the person who
connected them; a visitor never touches anyone's personal account. Keys for
custom tools or your own AI provider are encrypted and never shown again.
Deleting an agent, a document or a key is immediate and cannot be undone;
copies left in our backups are removed as the backups roll over, within 30 days.
Answers are written by AI and can be wrong. Terms: https://agents.clearmakes.com/terms
Privacy: https://agents.clearmakes.com/privacy  Security: https://agents.clearmakes.com/security