VegaDūtaAI · Agents
AI Agents on VegaDūta
An AI agent on VegaDūta is closer to a staff member than a chatbot: you give it a role, tools, and channels, and it does real work. Each agent runs on a Java/Spring runtime with more than 50 built-in tools — web fetch, document generation, image and video generation, knowledge search, memory, code execution, and more — and you deploy the same agent to WhatsApp, Telegram, Slack, Discord, Instagram, Messenger, X, email, or voice.
Guardrails and human-approval gates sit in front of risky actions, so the agent asks you first instead of acting unilaterally. Agents can also work as a team, with a lead agent delegating to specialists in parallel, and you can hand anyone a public share link to chat with an agent without creating an account.
By the VegaDūta team · Last updated
In short
- An AI agent on VegaDūta is a role, a set of tools and one or more channels: it carries out work rather than only answering chat messages.
- The same agent can be deployed to WhatsApp, Telegram, Slack, Discord, Instagram, Messenger, X, email and voice.
- VegaDūta agents ship with more than 50 built-in tools and 14 built-in Agent Skills, and a lead agent can delegate to a team of specialist agents in parallel.
- Guardrails and human-approval gates sit in front of risky actions, so an agent asks before it acts.
What an agent can actually do
Tools are what separate an agent from a chat window. VegaDūta agents ship with more than 50 built-in tools spanning research, content, memory, and action, and you extend the set further through MCP servers.
- Web fetch, document generation, image and video generation
- Knowledge search over your uploaded documents, plus memory recall and remember
- Google Workspace suite, Sarvam language suite, sandboxed code execution with your own E2B key
- Telegram send, speak (text-to-speech), code intelligence, delegate-to-agent
One agent, every channel
You build the agent once and attach channels to it. The same agent answers on WhatsApp, Telegram, Slack, Discord, Instagram, Facebook Messenger, X, email, and voice — no per-channel rebuild. This matters in India, where the customer conversation usually starts on WhatsApp, not on your website.
It asks you first
Sensitive tool calls can be gated behind human approval: the agent proposes the action, a person approves or rejects it, and the platform logs the decision. Combined with built-in guardrails for PII, toxicity, and jailbreak attempts, this keeps an autonomous agent inside boundaries you set. Sampled LLM-as-judge quality scoring gives you an ongoing read on how well each agent is performing.
Agents that work together
Delegation is itself a tool, so one agent can hand a subtask to a specialist agent and use the result. Add two or more members to an agent's team and the lead can delegate to several of them in parallel and combine the answers. Delegation goes one level deep, each call runs at most 8 branches, each branch is stopped after 240 seconds, and all branches share one per-run budget.
Skills an agent loads on demand
An Agent Skill is a packaged set of instructions for one kind of task, stored as a folder with a SKILL.md file. The agent sees only each skill's name and description until a request matches, then loads the full instructions. Fourteen skills are built in, covering documents (docx, pdf, pptx, xlsx), writing, code review, data analysis, and SQL, and a workspace admin can add custom ones.
Testing and versions
Every saved change to an agent is a version that can be restored, with staging and production labels. The Agent Lab in the playground draws each turn as a run graph, lets you fork the conversation from any message, and tests which sections of the prompt matter. Eval datasets and a promotion gate keep a failing version out of production.
Frequently asked questions
What is an AI agent, in plain language?
An AI agent is a language model given a role, tools, and permission to act — it can look things up, generate documents, remember facts, and message people, not just reply in a chat box. On VegaDūta you configure the role and tools, attach channels like WhatsApp, and the agent handles conversations and tasks on them.
Which channels can a VegaDūta agent work on?
WhatsApp, Telegram, Slack, Discord, Instagram, Facebook Messenger, X (Twitter), email, and voice. You attach channels to an agent rather than rebuilding it per channel, so one agent can cover all of them at once.
Can I stop an agent from doing something risky?
Yes. Risky tool calls can require human approval — the agent asks first, you approve or reject, and the decision is audit-logged. Guardrails for PII, toxicity, and jailbreak detection run on top of that.
Can several agents work on one task together?
Yes. An agent with two or more team members can delegate parts of a task to them in parallel and combine the results. Delegation is one level deep and bounded by a branch cap, a per-branch time limit, and a shared per-run budget.
Can people talk to my agent without an account?
Yes. Every agent can have a public share link that opens a chat with it — no sign-up required for the visitor. There is also a no-signup sandbox at /try where you can experience the platform itself before creating an account.
Sources
See it working in two minutes
The sandbox provisions a real tenant — describe an agent in one sentence and test it, no account, no card. Or browse ~90 industry workflow recipes to see what teams build.
