How long does it take to create a Babuba agent?

Under five seconds. One command — Babuba writes the agent's identity (SOUL), seeds its memory, wires its tools and channels and boots it live on your gateway. No provisioning, no deployments.

Are there limits on how many agents I can create?

No. Unlimited agents — one per department, project or experiment — with no per-agent fees and no approval needed to add one.

Can I change an agent after it is live?

Yes, with one command at runtime — edit its SOUL, swap its model, toggle its tools, re-scope its users. No redeploy.

Where do my agents run?

On your own server. Babuba is a self-hosted AI agent gateway — agents, memory and data stay on your hardware.

Babuba Agents — Create one in seconds. No per-seat fees.

Q: Does Babuba charge per seat?

No. Every agent serves the whole team — assign admins and users by phone or Telegram ID with zero seating costs.

gateway — create agent

$ babuba agents create --name finance --dept accounts

▸ generating SOUL.md ........ done (0.4s)

▸ seeding MEMORY.md ......... done (0.2s)

▸ binding tools (12) ....... done

▸ attaching channels ...... WA · TG · GM · WEB

▸ agent online ............ RUNNING // total elapsed 2.1s

$ babuba agents add-user --agent finance --user asha --role admin

▸ user asha scoped to finance . done

CREATE

<5 SEC

Agent live. Identity, memory, tools and channels wired automatically. You only name it and pick a department.

AGENTS

∞

Unlimited. One per department, one per project, one per experiment — no per-agent fees, no approvals to add one.

USERS / AGENT

∞

Every agent serves the whole team. Assign admins and users by phone or Telegram ID — zero seating costs.

REPLACE

1 CMD

Edit an agent's SOUL, swap its model, toggle its tools, re-scope its users — all at runtime, no redeploy.

// ONE AGENT PER DEPARTMENT

ISOLATED. SCOPED. AUDITABLE.

Each agent owns its memory, tools, guardrails and approval rules. Finance can't see HR's data. Ops can't approve payments. The router sends every message to the right agent — automatically.

FIN.ACCOUNTS RUNNING

Invoice capture & validation
GST reconciliation vs filings
Approval gates on every payment

$ 47 invoices matched today

OPS.LOGISTICS RUNNING

Dispatch tracking, live
Freight billing, auto-checked
Pending pickups chased

$ 12 dispatches verified

HR.PEOPLE RUNNING

Leave & attendance handling
Offer letters, generated
Onboarding checklists, tracked

$ payroll pack ready

ADMIN.COMPLIANCE RUNNING

Contract & renewal reminders
Regulatory deadlines tracked
Audit trail, searchable

$ 3 deadlines in next 30 days

// MULTI-USER BY DESIGN

EVERY AGENT. THE WHOLE TEAM.

Agents aren't single-user toys. Add your team in one command — each person gets scoped access, an audit trail, and their own view of the agent.

ROLE	LEVEL	WHAT THEY CAN DO	SCOPE
SUPER ADMIN	5	Everything — agents, RBAC, config, approvals, channels	ALL AGENTS
ADMIN	4	Manage agents & users in their department, run workflows	DEPARTMENT
MANAGER	3	Trigger workflows, approve actions, view reports	DEPARTMENT
USER	2	Chat with agents, request work, get results	ASSIGNED AGENT
VIEWER	1	Read-only — dashboards, logs, reports	ASSIGNED AGENT

Custom roles too — define your own level and permissions. Assign by phone or Telegram ID. Revoke in one tap. Everything logged.

// GUARDRAILS & APPROVALS

AGENTS PROPOSE. PEOPLE DECIDE.

Every agent ships with guardrails and approval gates. High-risk actions — payments, exports, sends — land on a human's phone with the full picture. One tap to approve, one tap to reject.

CREATE

MAKE YOUR OWN

One command — an agent with identity, memory, tools and channels, live in seconds.

$ babuba agents create --name finance

01

GUARDRAIL

Per-agent rules: allow, deny, warn or require-approval per topic. Enforced before and after every model call.

$ guardrail add --deny "salary data"

02

PREPARE

Agent prepares and verifies the action — invoice, export, email — with the full context attached.

$ payment 1044 --verify

03

APPROVE

The action lands on your phone. One tap. Approve or reject with a note — audit-logged either way.

$ await human.approval

04

SHIP

Approved actions execute with a full audit trail — who, what, when, which rule allowed it.

$ audit.log --tail

// LIVE EXAMPLE

WATCH IT WORK.

A real request, end to end — user asks, agent works, human approves.

fin.accounts — live request

[in] asha: "reconcile INV-1044, why is it pending?" ← whatsapp

[route] intent=reconciliation → fin.accounts

[work] parse · validate · match GST 47/47 matched

[flag] INV-1044 freight mismatch ₹3,200 → approval gate

[out] "Matched 47/47. INV-1044 flagged. Approve adjustment?" → whatsapp

[tap] asha approves → done · audit logged

// HOW THE ROUTER WORKS

ONE TASK. PARALLEL AGENTS.

Incoming message → intent detected → routed to the right agents working in parallel, each one showing its work before results merge.

ROUTER ▸ SCORES INTENT ▸ PICKS ONE AGENT ▸ DISPATCH

CREATE AN AGENT.
IN SECONDS.