New Connect any REST API as an approval-gated tool

Build, ground & ship
AI agents your users trust

Vaani is the self-serve platform for production AI agents. Give them your knowledge, turn any API into a typed, approval-gated tool, and deploy on one multi-tenant runtime that isolates every end-user. No fork, ever.

Runs on Google ADK · Gemini · MCP — swap models without rewrites
Grounded
12 docs in KB
Isolated
per end-user
app.vaani.ai · Support Agent Live
Book me a cardiology follow-up for next Tuesday afternoon.
Sure — Dr. Rao has 3:30pm open on Tuesday. I'll need your confirmation before I book it.
gated appointments.createcalendar.find_slot
Approval required — write action
create appointment · patient pinned server-side · nothing is committed until you approve
The runtime, models & protocols under the hood
Google ADK Gemini LiteLLM MCP OpenAPI Postgres Chroma
Everything in one platform

From prompt to production
without the plumbing

Configure behaviour, ground it in your data, wire up real tools, and ship — all from a single console.

Visual builder

Stand up an agent in minutes

A guided wizard walks you from a blank page to a live agent — persona, model tier, and policy, no YAML or redeploys.

  • Pick a Gemini Flash or Pro tier — or bring your own via LiteLLM
  • Test the draft live in a side-by-side chat before you publish
  • Every agent is config — the runtime never forks
Open the builder
Quick setupSupport Assistant
BehaviourGemini 2.5 Flash
3Knowledge base12 docs
4Connected tools2 APIs
5Deploymint key
Retrieval-grounded

Answers grounded in your knowledge

Upload markdown, text, or PDFs and Vaani ingests them into the agent's own private collection — retrieved at answer time to cut hallucinations.

  • Per-agent vector collections — knowledge never leaks across tenants
  • Citations link every answer back to its source
  • Long-term memory distils durable facts across sessions
refund-policy.pdf
240 KB · 18 chunks
Ready
onboarding.md
31 KB · 6 chunks
Ready
“What's our refund window for annual plans?”
Annual plans can be refunded within 30 days of renewal…
refund-policy.pdf · p.2
Tools & connectors

Every API becomes a safe tool

Import an OpenAPI spec and Vaani generates one typed, approval-gated tool per endpoint. Connect MCP servers too — reads flow freely, writes pause for a human.

  • One typed tool per endpoint — params validated automatically
  • End-user identity pinned server-side — no cross-user data access
  • Encrypted secrets, never exposed to the model
GET/orders/{id}
GET/customers/{id}/tickets
POST/refundsgated
DEL/subscriptions/{id}gated
Governance by config

Policy & isolation, built in

The same runtime serves a casual tutor and a regulated hospital — the difference is a policy profile, not a code fork. Tune autonomy, memory, audit, and residency per agent.

  • Human approval gates on every write action
  • Immutable audit trail & per-action auth strength
  • One API key, many end-users — fully isolated
Autonomy
act on gated tools directly
ActPropose
Memory
cross-session recall
RichMinimalNone
Audit
trail of every action
BasicImmutable
Residency
where state lives
LocalRegion
From zero to live

Ship in three steps

Configure in the console, connect what your agent needs, then call one endpoint from your app.

1

Configure

Sign up, describe the persona, choose a model and a policy profile. Preview it live — no key needed.

2

Connect

Upload knowledge, import an OpenAPI spec or MCP server, and store encrypted secrets the model never sees.

3

Deploy

Mint a server API key and call /v1/chat with a per-request end-user ID. One key, every user, isolated.

integrate.sh
# One key serves every end-user — Vaani isolates them server-side
curl https://api.vaani.ai/v1/chat \
  -H "Authorization: Bearer sk_live_•••••" \
  -H "Content-Type: application/json" \
  -d '{ "end_user_id": "user_42",
       "message": "Quiz me on Fundamental Rights" }'

# → { "session_id": "sess_…", "reply": "…",
#     "state_summary": { "mastery_changed": ["polity.fundamental_rights"] } }
One runtime, no forks

Built to scale to every tenant

1
Stateless runtime for every tenant
0
Forks to onboard a new client
End-users per API key, isolated
100%
Write actions behind a human gate
Loved by builders

Teams ship agents on Vaani

From edtech to regulated healthcare, one platform fits the policy to the use case.

★★★★★

“We went from a prototype notebook to a published, knowledge-grounded support agent in an afternoon. The approval gates were the unlock for our compliance team.”

RA
Riya Anand
Head of Support · fintech
★★★★★

“Importing our OpenAPI spec turned 40 endpoints into safe, typed tools instantly. End-user IDs pinned server-side meant we never worried about data leaks.”

DM
Daniel Mehta
Staff Engineer · logistics SaaS
★★★★★

“The same runtime runs our casual tutor and our regulated booking flow. We change a policy profile, not the codebase. That's a maintenance dream.”

SK
Sara Khan
CTO · health startup
Made for real work

So much you can build with Vaani

🎓Edtech

Adaptive tutor

RAG-grounded quizzes, mastery that persists across sessions, and proactive daily study nudges.

🏥Healthcare

Booking concierge

Books appointments behind a human approval gate — no PHI persisted, every action audited.

💬Support

Resolution agent

Answers from your docs, looks up orders, and issues refunds only after a human says yes.

Ship your first agent today

Sign up free, build in the console, and call one endpoint from your app. No fork, ever.