Build reliable AI agents faster, cheaper, consistently.

Kindgi is the full-stack agent framework: your coding agents build the agents, your team supervises, your experts decide, and Kindgi runs it all.

Faster Free to focus on your business.
Durable runs, approvals, versions and an audit trail come with Kindgi, so your coding agent never rebuilds them. Your first agent runs on your machine in about ten minutes, and any release rolls back in one step.
Cheaper Fewer AI tokens, in dev and in prod.
Using the Kindgi SDK, AI skills and AI-native docs, your coding agent writes only your agent's own logic. In production, flows and loops are plain code, and models are for judgment only.
Consistently The whole team builds one way.
From junior to senior, every developer's coding agent works from the same framework, templates, AI skills and docs. Every agent is built the same way, and the skills gap on your team narrows.

Keep scrolling to read more, or just see for yourself.

Set up Kindgi for me: read https://docs.kindgi.com/start/agent/ and follow it step by step. Tell me whenever you need me to do something, and wait for me.

Paste it into your coding agent.

Your people. Your knowledge. Your rules. All in Kindgi, the full-stack agent framework.

You bring

Kindgi brings

Your playbook: how your team does the work today

Instructions and flows: versioned prompts, steps, branches and retries

Your systems: the CRM, records and documents your team already works in

Typed tools: code, HTTP calls or MCP servers, checked on every call

Your rules: what the agent may do on its own, and what needs someone's sign-off

Approvals and guardrails: risky tool calls wait for a reviewer, and checks can halt a bad answer

Your expertise: your best people's judgment on real cases

Judgments and test sets: experts mark answers yes or no, and their verdicts become test cases

Your standards: the bar a new version has to clear before it goes live

Comparisons and release gates: a new version is compared on those cases, and a gate can hold it until it clears your bar

And Kindgi runs all of it on your own infrastructure, with:

the dev ergonomics of custom code
the flexibility of a plugin architecture
the guardrails of a framework
the audit trail of a compliance system
the durability of a workflow engine

On your machine in about ten minutes
In your test environments the same day
In production when your review signs off

Get started for free Talk to us

Beside your app. Inside your environment. Across your teams.Kindgi's anatomy at a glance.

Your environment

Developers and SREscoding agent and CLI

build and deploy

Domain expertsconsole or your own UI

judge answers, adjust instructions

Risk and complianceconsole

approve, audit

Your appany language

start runs, get results

Kindgi runtime a service you deploy

agentsflowsguardrailsapprovalsversions and evalsjournal and cost

runs your tools

Your packyour code, your repo

tool calls

Your systemsAPIs and MCP servers

writes every step

Your databaseruns, journal, versions

keeps

Secretsencrypted at rest

model calls, on your own accounts

Modelsyour provider account, your own GPUs, or your AI gateway

Your experts decide what good looks like.How your agents get better.

1
Your agent does real work.
It answers requests, drafts replies and updates records in your own systems.
2
Anything risky waits for a person.
You choose which actions need someone's approval before they happen.
3
Your experts mark answers right or wrong.
A yes or a no, with a reason if they like. Nobody has to write prompts.
4
Their judgments become the test.
Before release, a new version is tested on your real cases, and you can make that a rule.

Version 384%

Version 491%
Share of answers your experts approved. Example figures.
5
You decide what goes live.
Instructions change without waiting for engineering. Release when it's better, and roll back in one step.

Your business team owns what good looks like; engineering owns the code. Both work on the same agent, and nothing goes live without your say.

Built by your coding agent. Supervised by your people.How your agents get built.

Paste one prompt into Claude Code, Cursor, Codex or any coding agent. It sets up Kindgi, builds a first agent with you, and stops whenever it needs you.

Set up Kindgi for me: read https://docs.kindgi.com/start/agent/ and follow it step by step. Tell me whenever you need me to do something, and wait for me.

$ npx @kindgi/cli init my-pack --template=sample
$ cd my-pack && pnpm install
$ pnpm exec kindgi dev

Node 22.12 or later. Needs Docker.

$ uvx --from "kindgi-cli>=0.1,<0.2" kindgi init my-pack --template=python
$ cd my-pack && uv sync
$ uv run kindgi dev

Python 3.11 or later. Needs Docker.

$ npx @kindgi/cli init my-pack --template=java
$ cd my-pack
$ ./kindgiw dev

Java 17 or later; Maven comes with the pack. Needs Docker.

$ npx @kindgi/cli init my-pack --template=scala
$ cd my-pack
$ ./kindgiw dev

Scala 3.3 LTS or 2.13, on Java 17 or later, with sbt 1.10 or later. Needs Docker.

$ curl -X POST "$KINDGI_API_URL/v1/runs" \
    -H "Authorization: Bearer $KINDGI_API_TOKEN" \
    -H "Content-Type: application/json" \
    -d '{"agent":"my-pack.support","input":{"userMessage":"Where is order 1042?"}}'

Any language that speaks HTTP. Every route is in the OpenAPI spec. HTTP API reference

Kindgi gives coding agents everything they need to keep you in charge without keeping you busy.

Skills in your repo
Step-by-step instructions for tools, agents, flows and guardrails, in your language and matched to your Kindgi version.
Four ways in
The CLI, the SDKs and the HTTP API for your coding agent and your developers; the console for your business people.
Docs written for agents
Every detail documented in sync with Kindgi's interface, and all of docs.kindgi.com optimized for agent-to-agent communication.
Errors that name the fix
kindgi doctor checks the machine, and every error says what's wrong, so the agent fixes it and runs again.
You stay in charge
It builds and tests; your people decide what goes live. A release can require your experts' tests and a reviewer's sign-off, and each one is kept with who asked for it and why.
When Kindgi is the problem
Your agent tells a bug in Kindgi from a bug in your code, and writes it up, ready to send to us.

Running on your terms.Your knowledge, your decisions, your infrastructure, your models, a plain licence.

Where it runs

In your environment, as one more of your services: a container image and a database, inside your network, next to the services it works with.

Or on Kindgi Cloud, where we run it for you, in private preview.

Ask about Kindgi Cloud

Which models

Your agents don't name a model. They say what they need, and Kindgi picks from the models you connect: the leading model providers, open models you serve yourself, or your company's AI gateway.

Model calls use your own accounts.

Licence

The SDKs and CLI are open source (Apache-2.0). The runtime is free for development, testing and staging; production use needs a licence.

Licensing

Give Kindgi a try.On your own, or with us.

Get started for free

One prompt in your coding agent, and about ten minutes to a first agent on your machine.

A 4-week pilot

Kindgi in your environment, on one of your workflows, with weekly sessions with the founding team.

Apply

Talk to us

Thirty minutes on whether it fits your stack and your teams.

Book a call

We can't wait to see what you do with it.