Web SaaS · OSDU on-premise · AI-native

Kanysh.
Geology driven
by conversation.

An AI-native geological workstation. Say what you want — don’t think about how to do it in the menu. Seismic, wells, modeling — all in the browser, all through dialogue with the agent.

v 0.3· KZ · RU · EN· OSDU Data Platform· On-premise, no cloud
Request demo → Read the concept
Deployed inside corporate network· Data source — OSDU· Deployment — in a couple of hours· 100% on-premise · no telemetry

Say what you want —
don’t think how to do it in the menu.

The Kanysh Principle
§ 01 / Concept

Not yet another
Petrel replacement.
A different class of tool.

The desktop standard in oil & gas is built around 200+ buttons, menus and dialogs. Kanysh is built around the geologist’s intent. The control is one: chat and voice. Everything else — data, computation, rendering — runs on the server.

Aspect
Petrel · Kingdom
Kanysh
Control
200+ buttons, menus, dialogs, popups
Chat and voice. No toolbars or dialogs.
Data
Local files and projects on workstations
OSDU Data Platform. A single source of truth.
Compute
CPU on PC + a separate GPU module
Customer’s GPU server. The browser only renders.
Knowledge
Built-in help, PDF manuals
RAG over regulations and project history.
Access
Install and license per PC
Browser. Any device in the corporate network.
Permissions
File system, NTFS ACL
OSDU Entitlements + Keycloak SSO.
Language
English
Kazakh, Russian, English.
§ 02 / Interface

90% of the screen — data.
10% — conversation.

The left icon rail switches viewport modes. The right panel is the dialogue with the agent and task progress. Between them — seismic, wells, 3D scene. No menus — because they’re not needed.

Kanysh
проект: Tengiz-East  ·  OSDU: connected
● GPU · 71% VRAM · выделено geologist
⬡ ◈ ⌇ ▦ ≋ ⊞
inline 1247 · crossline 3204 tile · seismic/tengiz/1247.png z: t-domain · 0—5200 ms
Top Triassic · auto-tracked
SDMS stream CDP 4120 ────────────────── 4380 colormap: copper
Agent · LangGraph live
geologist
Show inline 1247 for Tengiz and track the Top Triassic.
→ osdu_get_seismic(survey="tengiz-east", inline=1247)
→ track_horizon(name="Top Triassic", method="seisnet")
kanysh
The slice is rendered. Top Triassic tracked over 92% of the area, uncertain zones highlighted — check near CDP 4280. Save the horizon to OSDU?
geologist
Compute Wyllie porosity for KZ-12 in this interval.
● _ ↵
Tengiz-East · OSDU · 4 скважины · очередь: 2 задачи online
§ 03 / Architecture

OSDU is the single source
of truth. GPU on the server.
Geologist in the browser.

Kanysh stores zero bytes of geophysical data. All reads and writes go through the OSDU Data Platform already deployed inside the corporate network. Seismic, wells, petrophysics, horizons — they live there. Permissions are inherited from Entitlements.

Client

Geologist’s browser
  • Three.js · WebGL · Canvas · WebSocket
  • Tiled seismic streaming — like Google Maps
  • Web Speech API + server-side fallback
  • No GPU on the PC required

Kanysh Server

Agent + GPU compute
  • Customer’s GPU server · scales with workload
  • FastAPI · WebSocket · LangGraph agent
  • LLM is configurable: vLLM on-prem (Qwen3-72B) or Anthropic / OpenAI API — via LiteLLM
  • CuPy · RAPIDS · PyTorch — deterministic computation
  • Qdrant — RAG over corporate documents
  • Celery — GPU task queue

Data source

OSDU Data Platform
  • Already running OSDU? We connect to it. If not, agents deploy it.
  • Seismic Store (SDMS) — slices without downloading SEG-Y
  • Well Delivery — LAS, markers
  • Search Service · File Service
  • Keycloak — SSO and Entitlements
  • Write results back to OSDU
BROWSER  ⟶  KANYSH AGENT  ⟶  OSDU SDMS / WELL DELIVERY  ⟶  GPU  ⟶  OSDU (write result)
§ 04 / How it works

The agent is an orchestrator. It doesn’t compute — it picks the tool.

The LLM understands intent and calls deterministic code on the GPU. Numbers always come from verified algorithms — no hallucinations in numerical results. Every value cites a regulation or an OSDU record.

  • 01 — IntentThe geologist speaks or types in natural language. Kazakh, Russian or English.
  • 02 — RouteThe agent picks a tool from the set: osdu_get_seismic, track_horizon, compute_petrophysics…
  • 03 — ComputeThe GPU server runs deterministic code. WebSocket streams progress.
  • 04 — SourceThe answer cites a regulation or an OSDU record. No source — no answer.
  • 05 — WriteThe result is saved back to OSDU. Colleagues see it instantly.
example · dialogue with the agent# Geologist: "Compute Wyllie porosity for KZ-12 in the Jurassic" # Agent: osdu_get_well_logs(well="KZ-12", curves=["DT","RHOB"]) ↳ LAS retrieved from OSDU Well Delivery compute_petrophysics( logs=lasdata, method="wyllie", interval="auto:jurassic" ) ↳ GPU · CuPy · 1.7s · deterministic osdu_save_property(name="Phi_KZ-12_J", ...) # Answer: KZ-12, Jurassic 3105 — 3248 m: Φ_avg = 14.7% method Wyllie · regulation § 7.3 p. 34 saved to OSDU · ID qsh:phi:4a82c
§ 05 / Capabilities

A full interpretation cycle.
No per-seat software to buy.

Each capability is a tool for the agent. The geologist doesn’t learn names — they just speak. Computation is deterministic, results are saved into OSDU.

Seismic

Cube navigation and processing
  • inline / crossline / timeslice
  • ML horizon tracking (SeisNet)
  • attributes: RMS, AI, instantaneous phase
  • filters: bandpass, f-k, deconvolution
  • tiled streaming from OSDU SDMS

Wells

Logging and correlation
  • LAS curve display
  • formation correlation (LAS-BERT)
  • petrophysics: Φ, Sw, Vsh
  • methods: Wyllie, Archie, Indonesia
  • stratigraphic markers from OSDU

Modeling

Structure and properties
  • time → depth conversion
  • structural model (horizons + faults)
  • geostatistics: SGS, Kriging
  • flow simulation: OPM Flow
  • export to RESQML

Analytics

Reserves and QC
  • volumetric reserve estimation
  • cross-plots on any curves
  • auto-QC of OSDU data
  • PDF reports · auto-generated

Knowledge · RAG

Corporate regulations
  • petrophysical methodologies
  • PRMS, log-analysis regulations
  • per-field interpretation history
  • source citation in every answer

Interface

Voice and split mode
  • text input · voice input
  • split viewport up to 4 panels
  • 3D scene: horizons + wells
  • Undo / Redo at the level of agent actions
  • export: LAS, CSV, PNG, RESQML
§ 06 / Deployment

A couple of hours.
On your servers.
Hands-free.

You grant access to your servers through the browser — the rest is done by AI agents. They deploy OSDU, bring up the LLM and GPU stack, configure Kanysh and launch the application. No manual DevOps, no weeks of installation.

01 · Access
You connect your servers via the web
You provide machine addresses and keys. Hardware can be anything — 1 GPU node or a cluster, sized to your workload. No ISO images, no installers.
↳ customer · ~10 minutes
02 · Install
AI agents deploy the entire stack
They deploy OSDU Data Platform, bring up the LLM with weights, configure GPU compute, wire up authentication and tile streaming. In parallel and autonomously.
↳ AI agents · ~2 hours
03 · Launch
The geologist opens the browser
The application is already running. The first request — “show seismic” — and the geologist sees data. Updates and patches are applied the same way by agents, without IT involvement.
↳ all set
kanysh-deploy · live tracerunning [00:00] kanysh init cluster: prod-01 · 2 nodes detected [00:03] agent.osdu provisioning Keycloak · Postgres · MinIO [00:18] agent.osdu Search · SDMS · Well Delivery ↑ up [00:27] agent.llm fetching weights · Qwen3-72B ▓▓▓▓▓▓▓░░ 72% [00:54] agent.gpu CUDA · RAPIDS · CuPy · resource pool ready [01:12] agent.llm vLLM serving on :8000 ↑ up [01:31] agent.kanysh binding to OSDU · loading tool registry ✓ [01:48] agent.kanysh health checks passed · TLS issued · WS open [02:04] kanysh ready · open https://kanysh.your-domain.local ✓ live
§ 07 / Roadmap

From PoC to full
release.

Четыре фазы. Десять месяцев от первого среза в браузере до полного цикла интерпретации в браузере — без установок, без лицензий, без обновлений.

Phase 01 · now
PoC
2 months

OSDU Keycloak + Search + SDMS. Basic browser viewer. 5 agent tools. Text input. RAG over corporate documents — ready. Criterion: “show inline 1247” — and the geologist sees seismic.

Phase 02
Alpha · MVP
3 months

Full P0. Voice input. Writing results back to OSDU. Stable WebSocket progress. Basic attributes and petrophysics.

Phase 03
Beta
3 months

Structural modeling, well correlation, split viewport, 3D scene. Burn-in on real projects.

Phase 04
Release
2 months

Geostatistics, basic flow simulation, PDF reports. Security review. Production release.

§ 08 / Investment

Open to product
investment.

Kanysh is a category, not a feature: an AI-native replacement for desktop geoscience workstations in a market with exactly two alternatives. We’re open to conversations with investors, strategic partners and industrial customers — especially those who understand OSDU, GPU infrastructure and oil & gas interpretation.

Get in touch →
§ 09 / FAQ

Frequently asked questions.

Quick answers to what people ask us first: how Kanysh is different, where the data lives, who deploys it.

+What is Kanysh?
Kanysh is an AI-native geological workstation. A conversational AI agent for geologists: seismic interpretation, well log analysis, petrophysics and reservoir modeling on top of the OSDU Data Platform. Runs in the browser, computes on the customer’s GPU server, on-premise.
+Is Kanysh an AI agent for geology?
Yes. The geologist speaks or types in natural language — Kazakh, Russian or English — and the agent orchestrates deterministic tools on the GPU. Numbers always come from verified algorithms, every value cites a regulation or an OSDU record. There are no hallucinations in numerical results.
+How does Kanysh compare to Petrel and Kingdom?
Petrel and Kingdom are desktop apps with 200+ buttons. Kanysh has no menus — only chat and voice. Data lives in OSDU instead of local files. Computation runs on a GPU server; the browser only renders. Access is via any browser inside the corporate network.
+Is Kanysh on-premise?
Fully. Kanysh is deployed inside the customer’s corporate network on the customer’s GPU servers. No cloud, no telemetry. AI agents automate the deployment of the entire stack — OSDU, LLM, GPU compute, Kanysh itself — in about two hours, without manual DevOps.
+What languages does the agent speak?
Kazakh, Russian, English. The geologist can switch freely mid-conversation — the agent keeps context regardless of language.
+We already run OSDU. Can Kanysh connect to it?
Yes. If you already have OSDU Data Platform — Kanysh connects to the existing instance via the standard APIs (Search Service, SDMS, Well Delivery, Entitlements). No data duplication, permissions are inherited from your Keycloak. If OSDU is not yet deployed — our AI agents bring it up as part of the install.
+Do I have to host the LLM on my own servers?
No, it’s configurable. By default we ship vLLM on-prem with the open Qwen3-72B — fully on-premise, data never leaves the perimeter. If your security policy allows it, you can route to API providers (Anthropic Claude, OpenAI GPT-4 and others) via LiteLLM. Seismic, well logs and petrophysics still live in OSDU on your servers — only text dialogue prompts go out.
+When can I see it?
PoC right now — basic viewer, RAG over corporate documents, 5 agent tools. Alpha with voice input and OSDU write-back arrives in 3 months. Full release in about 10 months. Request a demo.
Kanysh · v 0.3

Open the browser.
Type “show inline 1247”.
From there — work.

Request demo →