Kravos.ai / Agent Runtime Workbench

Compose production agents from real runtime parts.

KravosAI gives product and engineering teams one place to design, deploy, and operate agent behavior: knowledge, skills, tools, channels, analytics, and human fallback in the same control surface.

Build an agent stackRead the docs

Start with hosted runtime / keep API access from day one

Behavior

Versioned skills

Runtime

API + hosted UI

Fallback

Human takeover

Runtime Stack

01Knowledge
02Skills
03Tools
04Channels
05Fallback

How it works

A production path with fewer moving parts.

You still control the agent. Kravos.ai removes the repeat work around retrieval, channels, handoff, and operational visibility.

Removed from your backlog

Vector plumbinghandled
Channel forkshandled
Tool auth gluehandled
Takeover dashboardshandled
StepBuild note
01

Register the knowledge base

Connect URLs, sitemaps, files, and internal material. Kravos.ai handles ingestion, chunking, indexing, and retrieval plumbing.

02

Define what the agent may do

Attach tools, MCP servers, interceptors, escalation rules, and model settings per agent instead of burying behavior in glue code.

03

Deploy the same agent everywhere

Expose the agent through the API, web widget, WhatsApp, Messenger, or voice while preserving one conversation model.

04

Operate it after launch

Review conversations, takeover queues, usage, feedback, and channel health from the control plane your support team can actually use.

Skills

Reusable behavior modules, not copied prompt blocks.

Skills let teams package instructions once, attach them to agents, control runtime order, and keep the prompt footprint small until the agent needs the full instructions.

Runtime contract

Step

Attach

Step

Order

Step

Fetch

Agent runtime path

fetch_skill

01

Agent receives message

02

Retrieves tenant knowledge

03

Loads matching skill

04

Calls approved tools

05

Answers or escalates

Instruction

Support escalation policy

Decides when the agent should stop answering and move the thread to a person.

Behavior

Sales qualification

Collects fit, intent, budget, timeline, and routing details before syncing a lead.

Workflow

Internal helpdesk resolver

Uses source material and approved tools to answer policy, IT, and operations questions.

Capabilities

Use cases are where the feature list becomes honest.

Support Ops

Specimen

Answer customers from your own material, then hand off when the edge case appears.

For teams replacing first-line support without losing the human queue behind it.

QuestionRetrieveAnswerEscalate

Platform pieces

  • RAG ingestion
  • Widget + WhatsApp + Messenger
  • Sentiment and keyword escalation
  • Operator takeover

Sales + Success

Qualify, route, and follow up without making the agent a dead-end chatbot.

For revenue workflows that need facts, tool calls, and clean escalation paths.

LeadQualifyRouteSync

Platform pieces

  • Custom HTTP tools
  • Lead capture
  • Conversation memory
  • Webhook events

Internal Helpdesk

Put policy, runbooks, and system actions behind one conversational interface.

For internal teams tired of maintaining scattered Slack bots and brittle scripts.

AskCheckActLog

Platform pieces

  • Private knowledge sources
  • MCP servers
  • Role-based access
  • Audit-friendly activity

Voice Front Desk

Test and run voice agents without splitting the brain from your chat agent.

For appointment, triage, intake, and routing flows where the phone still matters.

CallListenResolveRecord

Platform pieces

  • ElevenLabs voice
  • Voice lab testing
  • Shared transcripts
  • Mid-call tool calls

Headless Product AI

Specimen

Embed agent behavior directly into your product with the same backend controls.

For developers who want an API first, not a UI-only automation toy.

POSTStreamCiteWebhook

Platform pieces

  • REST API
  • Streaming-ready runtime
  • React SDK
  • Per-key permissions

Operations

See what happened after launch: usage, channels, takeover load, and user feedback.

For teams that need to run agents like production software, not a prompt experiment.

ReviewMeasureTuneReport

Platform pieces

  • Analytics dashboard
  • Conversation review
  • Feedback signals
  • Tenant-scoped data

API

No black box between your product and the agent.

Use the UI when it helps. Use the API when the agent belongs inside your own application, workflow, or backend service.

Response contract

replyconversationIdcitationshandoffState
POST /v1/chat200 OK
1
// Create a conversation turn from your product
2
3
const response = await fetch('https://api.kravos.ai/v1/chat', {
4
method: 'POST',
5
headers: {
6
Authorization: 'Bearer sk_live_...',
7
'Content-Type': 'application/json',
8
},
9
body: JSON.stringify({
10
message: 'Can I reschedule my delivery?',
11
channel: 'web',
12
conversationId: 'conv_abc123', // optional — continues thread
13
}),
14
})
15
16
const { reply, conversationId, citations } = await response.json()

FAQ

Short answers before the docs.

Next build

Put an agent in front of real users without rebuilding the support stack.

Start with the hosted control plane, keep API access from day one, and bring humans into the loop when automation should stop.

Start freeOpen docs