Build, connect, and operate AI Agents

Recommended roadmap

1. 1.Create an AutronAI account or sign in.
2. 2.Prepare only the resources your use case needs: structured data, documents, Skills, or MCP servers.
3. 3.Create focused Agents and connect each Agent to the appropriate resources.
4. 4.Create a Chatbot and attach the Agents and Memory it should use.
5. 5.Review the composition in Chatbot Studio and verify instructions and access.
6. 6.Connect the Chatbot to the communication channels you need.
7. 7.Test real requests before wider deployment.
8. 8.If Automation is enabled for an eligible private Chatbot, review its schedule and runs.
9. 9.Monitor requests, latency, cost, errors, and delivery status.

Create and access your account

1. 1.Open Register and enter your name, email, and password.
2. 2.Complete the current verification step and create the account.
3. 3.Sign in with email/password, Google, or GitHub when available.

Navigation

• Overview: workspace status and operational summary.
• Data, Skills, and MCP Servers: reusable resources and capabilities.
• Agents: focused capabilities that can be attached to Chatbots.
• Memory and Chatbots: context and conversational composition.
• Integrations and Automations: deployment and scheduled work.
• Analytics, Request History, Wallet, and Support: operations and account management.

Initial setup

• Choose the correct language, theme, and time zone in account settings.
• Review notification settings before enabling alerts.
• Use the customer workspace for your own resources and configurations.

Structured Data

1. 1.Open Data Sources and create a structured source.
2. 2.Choose a supported database connection or structured file upload.
3. 3.Process or refresh the source so its schema is available.
4. 4.Select only the tables and columns the SQL Agent needs.
5. 5.Add descriptions where table or column meaning is not obvious.
6. 6.Save and verify the source status before using it.

Documents

1. 1.Create a Document source and enter a useful name and description.
2. 2.Upload a currently supported document type.
3. 3.Wait for processing to complete successfully.
4. 4.Use the document when configuring a Document Agent.

Data access

• Do not connect passwords, tokens, or unrelated sensitive data.
• Grant the smallest set of tables, columns, and documents needed for the use case.
• Treat a source as an Agent resource; it is not automatically available to every Chatbot.

Explore and install

1. 1.Open Skills and use Explore.
2. 2.Search or filter by category.
3. 3.Review a Skill's description, version, and instructions.
4. 4.Install it when it matches the intended workflow.
5. 5.Manage the installed copy from the Installed or My Skills area.

Create a Skill

1. 1.Open your personal Skills area and select Create Skill.
2. 2.Enter the slug, name, description, and focused instructions.
3. 3.Define the expected inputs, sequence, and outcome.
4. 4.Save the Skill and select it later when configuring an Agent.

Safe use

• Keep one Skill focused on one repeatable task.
• Do not place secrets or credentials in Skill instructions.
• Test missing inputs and failure paths before attaching the Skill to a live Chatbot.

Prepare

• Confirm the server address and provider.
• Identify its authentication method and required credentials.
• Know which tools and permissions the Agent actually needs.

Install and review

1. 1.Explore or create an MCP server in MCP Servers.
2. 2.Complete authentication and synchronize its tool list.
3. 3.Inspect new, changed, and removed tools.
4. 4.Approve only tools whose descriptions and access levels match expectations.

Attach selected tools

1. 1.Open Agents and create or edit a Tool Agent.
2. 2.Select the installed MCP instance.
3. 3.Select individual tools rather than granting the whole server by default.
4. 4.Attach Skills when they help the Agent use those tools correctly.
5. 5.Save and test the Agent with a safe request.

Create an Agent

1. 1.Open Agents and select Create Agent.
2. 2.Enter the name, slug, and concise description.
3. 3.Choose the Agent type.
4. 4.Write clear instructions and continue to Configuration.
5. 5.Select the resource required by that Agent type.
6. 6.Select relevant Skills and review the configuration.
7. 7.Create the Agent.

Agent types

• Custom Agent: focused instructions and capabilities for a specific role.
• SQL Agent: works with selected Structured Data.
• Document Agent: answers from selected Documents.
• Tool Agent: calls selected tools from an MCP server.

Design rules

• Give each Agent one clear responsibility.
• Do not put all data, tools, and workflows into one Agent.
• Keep instructions explicit about scope, refusal, and missing information.

Create Memory

1. 1.Open Memory and select Add or Create Memory.
2. 2.Enter a searchable title.
3. 3.Add the stable information that should be retained.
4. 4.Save the item and review its status.

Attach Memory to a Chatbot

1. 1.Open a Chatbot and enter Chatbot Studio.
2. 2.Use the Memory palette.
3. 3.Attach the required Memory item.
4. 4.Verify that it appears in the Chatbot composition and is enabled when supported.

Protect context

• Do not store passwords, API keys, tokens, or unnecessary personal data.
• Edit shared Memory carefully because several Chatbots may depend on it.
• Use Memory for durable context, not for a one-off task workflow.

Create a Chatbot

1. 1.Open Chatbots and create a Chatbot.
2. 2.Enter its name and configuration details.
3. 3.Open Chatbot Studio after creation.
4. 4.Review the available Agents and Memory in the palette.
5. 5.Attach only the capabilities this Chatbot should use.

Review the composition

• Select a node to inspect its details.
• Remove a node when the Chatbot no longer needs that capability.
• Check instructions, resource scope, Memory, and selected Agents before deployment.
• Keep Chatbot configuration separate from the focused logic inside each Agent.

Before deployment

1. 1.Test expected, incomplete, and out-of-scope requests.
2. 2.Confirm the Chatbot does not reveal internal data.
3. 3.Connect an Integration only after the composition behaves as expected.

General flow

1. 1.Open Integrations and select Create Integration.
2. 2.Choose the channel.
3. 3.Authenticate the provider or enter its Bot Token.
4. 4.Select the authenticated Page, OA, or Bot.
5. 5.Select the Chatbot and configure the integration fields.
6. 6.Complete the connection and test delivery from the real channel.

Provider connections

• Messenger and Zalo OA use the provider authorization flow.
• Telegram Bot and Zalo Bot require the provider's bot setup and token.
• Web Widget uses the generated embed snippet and its allowed-domain configuration.
• Never place tokens in tickets, notes, source code, or chat messages.

Verify delivery

1. 1.Confirm the Integration is active and points to the intended Chatbot.
2. 2.Send a test message from the actual channel.
3. 3.Verify the greeting, response, and error behavior.
4. 4.Revoke or regenerate credentials if they are exposed.

Create an Automation

1. 1.Open Automations and choose the create action.
2. 2.Select an eligible private Chatbot.
3. 3.Write the instruction and choose a one-time or recurring schedule.
4. 4.Set the timezone, deadline, and delivery configuration when available.
5. 5.Review the schedule and save it.

Monitor runs

• Check whether the Automation is active or paused.
• Review the next run and timezone.
• Open Runs to inspect state, result, response time, delivery, and error.
• Distinguish retryable failures from configuration or permission failures before changing the schedule.

Safe operation

Set up the team

1. 1.Open Settings and find the connection or handoff controls.
2. 2.Connect the supported staff channel.
3. 3.Add each staff member with a recognizable name.
4. 4.Send a private connection link to the assigned person.
5. 5.Disable or revoke access when a person leaves the role.

Track sessions

• Pending: no staff member has accepted the session.
• In progress: a staff member is handling it.
• Resolved: the session is complete.
• Expired: the session is no longer active.

Protect staff access

Analytics

1. 1.Open Analytics.
2. 2.Choose the relevant date range.
3. 3.Filter by Chatbot or platform when available.
4. 4.Compare request volume, cost, latency, and error trends.
5. 5.Use the trend to decide which resource, Agent, Integration, or Automation needs inspection.

Request History

1. 1.Open Request History.
2. 2.Search by request, content, or entity using the available search.
3. 3.Apply filters and inspect status, timing, model, and usage details.
4. 4.Compare an unusual request with a normal request.
5. 5.Trace the issue back to the responsible configuration layer.

Investigate safely

Wallet

• View current balance, deposits, usage cost, refunds, and adjustments.
• Filter transactions and compare the before and after balance.
• Review low-balance alerts in notification settings when available.

Settings

• Keep language, theme, timezone, and notification preferences current.
• Review connected accounts and staff access.
• Change settings only when you understand which Chatbots, Integrations, or Automations they affect.

Support

1. 1.Open Support and create a request.
2. 2.Identify the affected Chatbot, Agent, Data Source, MCP server, Integration, or Automation.
3. 3.Include the approximate time, reproduction steps, actual result, and expected result.
4. 4.Attach a sanitized screenshot when useful.
5. 5.Never send passwords, tokens, API keys, or full connection strings.

Answers are incomplete or incorrect

1. 1.Find the request in Request History.
2. 2.Check whether the Chatbot selected the expected Agent.
3. 3.Verify the Agent's source, allowed tables or columns, document status, selected Skills, and Memory.
4. 4.Review instructions for conflicts or missing scope.
5. 5.Retest with the smallest reproducible request.

A tool or channel fails

1. 1.Confirm the MCP server or Integration is active.
2. 2.Confirm credentials or OAuth are valid.
3. 3.For MCP, confirm the exact tool exists and is selected for the Tool Agent.
4. 4.For a channel, confirm it points to the correct Chatbot and the embed/token configuration is current.
5. 5.Open Support if the issue remains.

An Automation fails

1. 1.Confirm the Chatbot is eligible and Automation is enabled.
2. 2.Check whether the Automation is paused.
3. 3.Verify next run, timezone, deadline, and latest run state.
4. 4.Inspect the result, delivery state, and error before editing the instruction or schedule.

Safety principles

• Grant only the data and tools required for the use case.
• Do not share secrets in prompts, Memory, Skills, tickets, or logs.
• Test on an internal channel before wider deployment.
• Disable or revoke access when unusual behavior or unauthorized access is detected.