DocsAI Agents

Business

AI Agents

Deploy autonomous AI chatbots to Telegram, Slack, and Email. Give them a knowledge base, memory, and custom tools — without writing code.

AI Agents require the Business plan. You can create 1 agent per account.

What Agents Can Do

AI Agents are autonomous chatbots that live on messaging channels. They respond to incoming messages, maintain memory of past conversations, search a knowledge base to answer domain-specific questions, and can send proactive scheduled messages. Think of them as always-on team members that work across Telegram, Slack, and Email.

Business Plan Required

AI Agents are exclusively available on the Business plan. You can create 1 agent per account. Ensure your plan is active before configuring channels or deploying.

Respond to messages

Receive and reply to messages on Telegram, Slack, or via email with AI-powered responses using your configured personality and instructions.

Knowledge base

Upload PDF, TXT, MD, and CSV files. Content is chunked and indexed in a vector DB — the agent retrieves relevant chunks for every response.

Memory

Episodic memory persists facts across conversations. The agent automatically stores and recalls information about users and topics via semantic search.

Scheduled messages

Cron-based proactive messages sent to configured channels at specific times. Supports quiet hours and timezone configuration.

Custom tools

Define HTTP GET/POST endpoints the agent can call. Supports header and body templates with {{variable}} interpolation. Up to 10 custom tools.

Built-in skills

web_search, http_request, current_time, and weather — four skills available out of the box. Enable or disable each per agent.

Human escalation

Detect trigger keywords like "speak to a human" and route to a human. Notifies via email or Slack with the conversation context.

Multi-channel support

Connect multiple channels to a single agent. Separate conversation history is maintained per participant across Telegram, Slack, and Email.

Setting Up an Agent

Creating and deploying an agent takes a few steps. The AI generates a personality and instruction set from your description — you can refine it before deploying.

Create a new agent

Navigate to the Agents page and click Create Agent. Give it a name and describe what it should do. The AI will generate an initial personality, tone, and instruction set based on your description.

Review AI-generated configuration

The AI generates a display name, tone, system instructions, and suggested boundaries. Review and edit any of these to match your intended use case before deploying.

Connect a channel

Add at least one channel — Telegram (bot token from @BotFather), Slack (channel IDs via OAuth), or Email (Gmail-based, optional trigger email). The agent starts responding to messages on connected channels immediately after deploy.

Add a knowledge base (optional)

Upload files the agent should reference when answering questions. PDF, TXT, MD, and CSV files are supported. Content is automatically chunked and indexed in a vector database.

Enable skills and tools

Turn on built-in skills (web_search, weather, etc.) as needed. Optionally define custom HTTP tools so the agent can call your own APIs or trigger external actions.

Deploy

Save your configuration. The agent is deployed to a dedicated GKE pod and starts responding to messages on all connected channels.

Start with a focused scope

Agents perform best when given a clear, specific purpose. A customer support agent for a SaaS product with a well-structured knowledge base will outperform a general-purpose assistant with no knowledge context.

Supported Channels

Agents operate on messaging channels. Each channel has its own configuration requirements. You can connect multiple channels to the same agent — the agent maintains separate conversation history per channel participant.

Bot token from @BotFather

Create a Telegram bot via @BotFather and paste the bot token into your agent's channel settings. The agent responds to any direct message or group mention. Supports text, and can send formatted messages with markdown.

Slack

Channel IDs via OAuth

Connect via OAuth and specify which Slack channel IDs the agent monitors and responds in. The agent can both respond to messages and send proactive scheduled messages to configured channels.

Email

Gmail-based, optional trigger email

Connect a Gmail account for email-based conversations. Optionally configure a trigger email address — messages forwarded to this address are routed to the agent. The agent replies directly to the sender.

You can connect multiple channels to a single agent. Conversation history is tracked per channel and per participant — a Telegram user and a Slack user interacting with the same agent have independent conversation histories.

Knowledge Base & Memory

Agents have two distinct ways of storing and recalling information: a static knowledge base for domain content you upload, and a dynamic memory system that automatically captures facts from conversations.

Knowledge Base

Upload files that the agent references when answering questions. Each file is chunked into overlapping segments and stored in a vector database. When a user asks a question, the agent runs a semantic search across all chunks and injects the most relevant results into its context before generating a response.

Supported formats	PDF, TXT, MD, CSV
Max file size	10MB per file
Upload rate limit	30 files per hour
Similarity threshold	Configurable — controls how closely a chunk must match the query
Result count	Configurable — number of chunks injected per response

Memory

Memory enables the agent to remember facts across separate conversations. The agent automatically extracts and stores memorable facts during interactions — user preferences, names, past topics, and other relevant details. When a new conversation starts, memories are recalled via semantic search and injected into the agent's context.

Storage	Up to 100 memories per agent (configurable)
Recall method	Semantic search against new conversation context
Scope	Per agent — memories are shared across all channels
Management	View and delete individual memories from the agent settings

Knowledge base vs. Memory

Use the knowledge base for stable reference content — product documentation, FAQs, policies. Use memory for dynamic, user-specific facts that emerge during conversations. They complement each other: knowledge base answers the question, memory personalizes the response.

Skills & Custom Tools

Skills and tools let the agent take action beyond generating text. Built-in skills cover common needs. Custom tools let you connect the agent to your own APIs and services.

Built-in Skills

Skill	Description
web_search	Performs a real-time web search and returns summarized results. Use for questions about current events, live data, or anything outside the knowledge base.
http_request	Makes an HTTP request to any URL and returns the response. Useful for fetching live API data during a conversation.
current_time	Returns the current date and time in any timezone. Useful for scheduling-related conversations.
weather	Returns current weather conditions for any location. Requires the location to be specified in the query.

Custom Tools

Define HTTP endpoints the agent can call during a conversation. Each tool has a name, description (used by the AI to decide when to invoke it), method (GET or POST), URL, optional headers, and an optional body template. Body templates support {{variable}} interpolation from the conversation context.

Method	GET or POST
URL	Any HTTPS endpoint
Headers	Key-value pairs (e.g., Authorization: Bearer {{token}})
Body template	JSON template with {{variable}} placeholders
Max tools	10 per agent

Escalation & Scheduled Tasks

Human Escalation

Configure keywords that trigger a handoff to a human. When the agent detects a configured keyword (e.g., "speak to a human", "urgent", "I need help now"), it sends a configurable escalation message to the user and notifies a human via email or Slack with the full conversation context.

Trigger keywords	Comma-separated list (e.g., "speak to a human, urgent, talk to support")
Escalation message	What the agent tells the user when escalating
Notification target	Email address or Slack channel ID to notify
Context included	Full conversation history sent with the notification

Scheduled Tasks

Scheduled tasks let the agent proactively send messages to a configured channel on a cron schedule. Each task has a cron expression, a target channel, and a message template. Quiet hours prevent messages from being sent outside business hours.

Max tasks	5 per agent
Schedule format	Standard cron expression (e.g., 0 9 * * 1-5 for weekday mornings)
Quiet hours	Start time, end time, and timezone — messages are suppressed in this window
Target channel	Any connected channel (Telegram, Slack, or Email)
Message template	Plain text with optional dynamic content the agent generates at send time

Use scheduled tasks for daily digests

Scheduled tasks are ideal for recurring updates — daily summaries, weekly reports, or morning briefings sent to a Slack channel. Combine with custom tools to fetch live data at send time.

Configuration Reference

Every agent setting and what it controls.

Setting	Description
Display name	The name your agent uses when introducing itself
Tone	Personality style: professional, friendly, concise, empathetic, etc.
Instructions	System-level instructions defining the agent's role and behavior
Boundaries	Topics the agent should not discuss
Values	Core principles to reflect in all responses
Skills	Enable/disable: web_search, http_request, current_time, weather
Custom tools	HTTP GET/POST endpoints with name, description, URL, headers, body template
Knowledge base	Uploaded files (PDF, TXT, MD, CSV) indexed for semantic search
Memory	Enable/disable episodic memory; configure max stored memories
Scheduled tasks	Cron expression + target channel + message template (max 5)
Quiet hours	Start/end time + timezone for suppressing scheduled messages
Escalation keywords	Comma-separated trigger phrases for human handoff
Escalation message	What the agent tells the user when escalating
Escalation target	Email or Slack channel ID to notify on escalation

Limits

Agents have usage limits to ensure reliable performance for all accounts.

Limit	Value
Agents per account	1
Knowledge base — max file size	10MB per file
Knowledge base — upload rate	30 files per hour
Memory — max entries	100 (configurable)
Scheduled tasks	5 per agent
Built-in skills	4 (web_search, http_request, current_time, weather)
Custom tools	10 per agent
Channels	Telegram, Slack, Email
Run model	Long-running GKE pod — no per-request timeout

FAQs

What is an AI Agent?▾

An AI Agent is an autonomous chatbot that lives on messaging channels like Telegram, Slack, and Email. It responds to incoming messages, maintains memory of past conversations, can proactively send scheduled messages, and can use tools to fetch live data or trigger external actions — all without any code.

How is an agent different from a regular app?▾

Apps are user-facing web applications deployed to a URL. Agents are channel-based — they run continuously in the background, respond to messages on Telegram, Slack, or Email, and can initiate conversations on a schedule. You interact with an agent through a messaging platform, not a browser.

Can the agent remember past conversations?▾

Yes. The Memory feature stores facts about users and topics automatically across conversations. Up to 100 memories per agent (configurable), recalled via semantic search. This is separate from the in-conversation context window — memory persists between sessions.

What can the knowledge base contain?▾

The knowledge base accepts PDF, TXT, MD, and CSV files up to 10MB each. Content is automatically chunked and stored in a vector database for semantic search. When a user asks a question, the agent retrieves the most relevant chunks and uses them as context. Upload rate is limited to 30 files per hour.

Can the agent send proactive messages?▾

Yes. Configure scheduled tasks with cron expressions to have the agent send messages to specific channels at specific times. You can define up to 5 scheduled tasks per agent, each with their own cron schedule, target channel, and message template. Quiet hours are supported to prevent unwanted notifications.

What built-in skills does the agent have?▾

Four built-in skills: web_search (real-time web search), http_request (call any URL), current_time (date/time in any timezone), and weather (conditions for any location). Each skill can be enabled or disabled per agent in the configuration.

What is escalation?▾

Escalation routes a conversation to a human when the agent detects certain keywords — for example, "speak to a human", "talk to support", or "urgent help". You configure the keywords to watch for, the escalation message the agent sends to the user, and where to send the notification (email address or Slack channel).

Ready to build?

Create your first app for free — no credit card required.

Embed Agent Mode