Bots

Aerostack Bots are a full agent builder, not just a chatbot. Every bot connects to one MCP workspace and gets instant access to all its tools — databases, payment APIs, ticketing systems, knowledge bases, and anything else you wire up. Two execution modes let you choose exactly how messages are handled: let the LLM figure it out autonomously, or design a deterministic workflow graph that runs exactly the steps you define.

Architecture

Messages arrive from any supported platform, get normalized into a unified format, then enter one of two execution modes. Both modes have full access to every tool in the bot’s workspace.

Two Execution Modes

Agent Loop Mode

The LLM autonomously decides which tools to call, in what order, and when to stop. Best for open-ended conversation, Q&A, and exploratory tasks where you cannot predict every path in advance.

• Up to 10 tool-calling loops per message (configurable)
• LLM sees all available tools from the workspace and picks the right ones
• Simple messages that need no tools are answered immediately (no tool overhead)
• Works with any LLM provider that supports tool calling

Workflow Mode

You design a directed graph of nodes and edges. The engine executes them in order, following your branching logic. Best for deterministic processes where every step must happen in a defined sequence.

• Up to 50 nodes per execution with cycle detection
• 14 node types: trigger, llm_call, logic, mcp_tool, send_message, action, loop, code_block, auth_gate, schedule_message, send_proactive, delegate_to_bot, error_handler, parallel
• Variable interpolation across all nodes
• LLM calls available at specific points (structured flow with AI reasoning at key steps)

Supported Platforms

LLM Provider Support

You can use Aerostack’s pooled API keys (billed via wallet) or bring your own key (BYOK mode, no platform charges).

Key Capabilities

Workspace Integration

Every bot connects to one MCP workspace. The workspace aggregates tools from multiple MCP servers, Skills, and Functions. Add a Stripe MCP to your workspace and the bot can immediately process payments. Add a database MCP and it can query your data. No code changes needed — the workspace is the single source of capabilities.

Auth Gate (Identity Verification)

The auth_gate workflow node runs a multi-turn identity verification flow directly inside the conversation. The bot prompts the user for their email or phone, sends a one-time code, and verifies it before proceeding. Four built-in providers (Resend, SES, Twilio, MSG91) plus custom HTTP. State machine: awaiting_contact to awaiting_otp to verified or failed.

Human Handoffs

A workflow can pause execution and notify a human reviewer via email, web dashboard, Telegram, or Discord. The reviewer sees the full conversation context, the reason for escalation, and approve/reject buttons. On approval, the workflow resumes down the approved path. On rejection, it follows the rejected path. See Human Handoffs for the full guide.

Bot-to-Bot Delegation

The delegate_to_bot node lets one bot call another bot. Each bot can live in its own workspace with its own tools. A reception bot can route to specialist bots (billing-bot, tech-bot, hr-bot) based on intent classification. Maximum 3 delegation hops.

Proactive and Scheduled Messages

Workflows can send messages to different channels or users (send_proactive) and schedule messages for future delivery (schedule_message). Use cases: appointment reminders, follow-up messages, shift notifications, daily digests.

Use Cases

Customer Support (Telegram)

A Telegram bot with a workflow that verifies the customer’s identity via auth_gate, looks up their order history via a database MCP, checks payment status via Stripe MCP, processes refunds for amounts under $500 automatically, and escalates higher amounts to a finance manager via human_handoff. The finance manager approves from their phone and the bot processes the refund.

Community Manager (Discord)

A Discord bot in agent loop mode connected to a workspace with a knowledge base MCP and a moderation MCP. Users ask questions with /ask and the bot searches the FAQ, answers from the knowledge base, and logs unanswered questions for the team to review. Onboards new members with a welcome workflow that collects their role and interests.

Sales Assistant (WhatsApp)

A WhatsApp bot that qualifies leads through a workflow: asks about company size, use case, and timeline. Scores the lead using a code_block node, then either schedules a demo via a calendar MCP or sends product materials. High-value leads trigger a human_handoff to the sales team with full context.

Internal Ops (Slack)

A Slack bot in a company workspace connected to MCPs for metrics dashboards, deployment systems, and incident management. Team members ask “What were yesterday’s error rates?” and the bot queries the observability MCP. Deployment requests go through a human_handoff to the on-call engineer.

HR Helpdesk (Slack)

A Slack bot using a workflow to handle PTO requests: verifies employee identity, checks remaining balance via an HR system MCP, submits the request, and routes approval to the employee’s manager via human_handoff. Also answers benefits questions using a knowledge base MCP in agent loop mode.

E-commerce Assistant (Telegram)

A Telegram bot for order tracking, returns, and product recommendations. Workflow mode for returns (verify identity, look up order, check return eligibility, process return or escalate). Agent loop mode for general product questions using a catalog MCP.

Appointment Scheduler (WhatsApp)

A WhatsApp bot that checks availability via a calendar MCP, books appointments, sends confirmation messages, and uses schedule_message to send reminders 24 hours and 1 hour before the appointment. Cancellation requests go through a simple workflow that frees the slot and confirms.

DevOps Incident Bot (Workflow)

A multi-bot setup: a monitoring bot receives alerts, parses severity using an llm_call node, and delegates to an oncall-bot via delegate_to_bot. The oncall-bot creates a PagerDuty incident via MCP, notifies the channel via send_proactive, and escalates P1 incidents to the engineering manager via human_handoff.

Aerostack vs Other Bot Builders

The key difference: Aerostack bots inherit capabilities from the MCP ecosystem. Any MCP server anyone publishes — open source or on the Aerostack marketplace — immediately becomes a capability your bot can use. No custom integration code.

Billing Modes

See Billing & Limits for pricing details, spending caps, and rate limits. See Cost Optimization for strategies to reduce costs.

Next Steps

• Create Your First Bot — Step-by-step dashboard tutorial
• Platform Guides — Set up Telegram, Discord, WhatsApp, Slack, or custom webhooks
• Connect Tools — Give your bot capabilities through workspaces
• Workflows — Design deterministic automation flows
• Human Handoffs — Pause for human approval
• Bot Teams & Delegation — Route between specialist bots