YantrikaOpen source · MIT licensed · v0.12.1
The operations layer for your robot fleet.
Yantrika is an open-source, end-to-end fleet-operations platform for multi-vendor AMR (autonomous mobile robot) fleets. Robots speak VDA 5050 v2.1 over MQTT; Yantrika gives you a live console, human-approved command gating, automatic incident detection, predictive maintenance, real alert delivery, and an AI copilot grounded in your fleet's actual data — self-hosted, on your own infrastructure.
Starts a private sandbox with its own simulated fleet and drops you straight into the console. Nothing to install, nothing to fill in. It expires on its own and takes its data with it.
10 automated test suites, 500+ tests, all passing — plus 188 more covering the conformance tester on its own. MIT license. Runs standalone or against your own Supabase/Postgres project.
Three commands, zero config
# Linux / macOS
$ git clone https://github.com/yantrika-ai/yantrika.git && cd yantrika
$ ./install.sh
$ .venv/bin/python -m yantraops up --loopbackThat's the full platform — simulated 10-AMR fleet, automatic incident detection, predictive maintenance, notifications, and the Sarathi AI copilot — running against an embedded in-memory backend. No cloud account, no API keys, no physical robots required.
Windows supported via install.ps1 and a PowerShell runbook.
What it does
One shared data layer, five cooperating services.
Robots — or Yantrika's own deterministic warehouse simulator — publish VDA 5050 v2.1 state messages over MQTT. A connector translates that standard protocol into a shared Postgres (Supabase) schema. Every other component — the console, the incident detector, the predictive-maintenance engine, the notifier, and the AI copilot — reads and writes that same schema, so what you see in the console is exactly what every other service is acting on.
Live fleet console
Single-file web console (no build step): live map, per-robot detail, alerts, incidents with recorded-telemetry replay, and a mission board.
Human approval gate
Robot commands (pause, send-to-charge, e-stop) don't execute directly — they queue as pending rows that an operator approves or rejects, with a full audit trail of who requested and who decided.
Automatic incident detection
A detector polls robot state and opens/resolves incidents with deterministic, idempotent IDs, flap-window re-opening, and stale auto-resolve — PagerDuty-style dedup logic, running against your own data.
Predictive maintenance
Reads recent telemetry windows and flags drive-motor temperature trends, abnormal battery drain, and drivetrain speed decline before they become failures — each finding carries a human-readable explanation, an estimated remaining-useful-life, and a confidence score.
Sarathi, the AI copilot
Ask questions about your fleet in plain language; answers are grounded in your fleet's live data with cited sources, and gracefully degrade from LLM-backed to template-based answers if no LLM key is configured — it always works, it just gets smarter with a key.
Built on an open standard
Yantrika doesn't lock you into one robot vendor. It speaks VDA 5050 v2.1, the open, vendor-neutral protocol for AMR fleet communication — so it's built to sit above multiple robot brands on the same floor, not just one. The bundled simulator models a 10-robot fleet across 3 simulated vendors on an 8×5 warehouse grid to prove this out without hardware.
A protocol only helps if the vehicles actually implement it. Yantrika ships a free VDA 5050 conformance tester — 52 checks, a graded HTML report, no account and no sales call — so you can find out before the robots are on your floor.
Deploy it your way
Self-hosted, on your own infrastructure.
Try it locally in minutes
install.sh / install.ps1 sets up a Python virtual environment and runs the full stack against an embedded in-memory backend. Zero cloud dependency.
Run it in Docker
docker build or docker compose up, host networking, Linux/CI-ready.
Deploy to your own Supabase/Postgres project
A numbered set of ordered, idempotent, checksummed SQL migrations under supabase/ brings up the schema; yantraops migrate applies and tracks them.
There is no hosted SaaS version. You run Yantrika on your own infrastructure, with your own data, under the MIT license.
Security & access control, built in
Yantrika ships in demo mode by default — one shared key, zero setup, so you can evaluate it in minutes. For real deployments, two opt-in migrations lock it down:
- Hardened mode — drops open read/write access; only authenticated sessions can read, and all writes move to a server-side service key.
- Full RBAC mode — role-based access control with four levels (
operator < engineer < manager < admin), scoped by site, enforced by Postgres row-level security policies andSECURITY DEFINERfunctions — not just application-layer checks. Command approval requires a manager-or-above role, going through a dedicated database function rather than a direct write path.
Every alert acknowledgement, every command request, and every approve/reject decision is attributed to a named person with a timestamp — a full audit trail out of the box.
Read more about the security model → · Full threat model & hardening guide (docs/SECURITY.md)
Common questions
Straight answers.
Is there a hosted SaaS version?
No. Yantrika is self-hosted software you run on your own infrastructure against your own database, under the MIT license. The live demo on this page is a throwaway sandbox that expires on its own; it is not a hosted product.
Does it work with robots from more than one vendor?
That is the point of it. Robots talk to Yantrika over VDA 5050 v2.1, the open vendor-neutral MQTT protocol for AMR fleet communication, so any vehicle that implements the standard can be driven from the same console. The bundled simulator models a 10-robot fleet across 3 simulated vendors to exercise that path without hardware.
Do I need an account or an email address to try it?
No. The Try it button mints a throwaway sandbox with its own site id and a small simulated fleet, and drops you straight into the console. It collects nothing about you, and the sandbox and its data are deleted when its time is up.
What does it cost?
Nothing. It is MIT licensed open-source software. There is no paid tier, no enterprise edition, and no feature held back behind a sales call. Your costs are whatever infrastructure you choose to run it on.
How do I check whether my robots actually implement VDA 5050 correctly?
Run yantra-conform, the conformance tester that ships with Yantrika. It grades a fleet against 52 checks drawn from VDA 5050 v2.1 and writes a single HTML report naming the clause, what was expected, what your vehicle actually did, and how to fix it. It is a command-line tool that runs on your machine against your broker, and it is free — here is what it tests and how to run it.
Yantrika is open source and self-hosted.
Clone it, run it against the bundled simulator in three commands, and see your first fleet dashboard today.