DOS AI

This repository holds the machine-readable contract and runnable examples. The platform itself is hosted, so there is nothing to install: you get an API key in the cabinet and call the endpoints below.

A project in DOS AI is one bot: its instructions, its knowledge base, its channel (WhatsApp via QR or Telegram), its balance and its clients. The bot answers incoming messages in natural language, books time slots from a Google Calendar or Sheet, issues invoices in the chat, files every conversation as a lead, and hands the dialog over to a human when your rules say so.

The API exposes that same data: projects, conversations, messages, leads, analytics and balance.

Every request carries an API key issued in the cabinet:

The key authenticates as the user who created it, so it inherits that user's access. Only a SHA-256 hash of the key is stored, which means a lost key cannot be recovered, only revoked and reissued. Keys can be scoped to a single project and to read-only access.

Rate limit: 120 requests per minute per key, plus a stricter per-endpoint budget. On 429 theRetry-Afterheader tells you how many seconds to wait.

export DOSAI_KEY="dos_sk_live_..." # projects visible to this key curl -s https://dosai.pro/api/projects \ -H "Authorization: Bearer $DOSAI_KEY" # leads of one project curl -s "https://dosai.pro/api/projects/$PROJECT_ID/leads?limit=20" \ -H "Authorization: Bearer $DOSAI_KEY" # analytics for a period curl -s "https://dosai.pro/api/projects/$PROJECT_ID/analytics?from=2026-08-01&to=2026-08-14" \ -H "Authorization: Bearer $DOSAI_KEY"

Available paths (full schemas in the OpenAPI file):

Responses come wrapped in an envelope ({ "data": ... }), so unwrap before use. The OpenAPI file is the authority here and is generated from the same source the routes are checked against.

The platform ships a Model Context Protocol endpoint, so an AI assistant can read and manage your bots directly. Point your MCP client at:

list_projects,get_project,get_prompt,update_prompt,list_functions,list_leads,get_lead,update_lead,list_conversations,get_messages,send_operator_message,get_analytics,get_balance.

No destructive tools.There is no delete-project, no payment, no member management. Adding one would be a conscious decision, not a reflex, and a test guards the list.

Every tool calls the public REST route over HTTP, not an internal service layer. That way a read-only key stays read-only, project scoping and rate limits apply exactly once, and the agent cannot reach anything a human with the same key could not.

Example client config (Claude Desktop style):

{ "mcpServers": { "dosai": { "url": "https://dosai.pro/api/mcp", "headers": { "Authorization": "Bearer dos_sk_live_..." } } } }

Examples in this repository are MIT. The platform itself is a hosted commercial service.

AI phone secretary for businesses: place outbound calls and read transcripts, leads, voicemails, SMS, and analytics from your business line.

Create AI-generated memes and convert them into stickers for Telegram and WhatsApp.

An MCP server integrating WhatsApp messaging and ElevenLabs AI voice capabilities into VS Code.

Build and publish production-ready AI agents for web, SMS, and phone directly from your coding agent via a remote MCP connection.

Configure and train your TBit WhatsApp/Instagram AI agents from ChatGPT or Claude (remote, OAuth 2.1)

Integrate AI assistants with VOYP's calling capabilities to make calls, search business info, and manage call status.

Connect your personal WhatsApp account to an AI agent using the WhatsApp Web multi-device API.

Connects a personal WhatsApp account to an AI agent using the WhatsApp Web multi-device API.

graspil.com is an analytics, broadcast, and automation platform for Telegram bots, channels, and groups. Through MCP, an agent can build reports and dashboards, manage broadcasts and automations, and view conversation history and referral stats — all without logging into the dashboard.