PL Book a demo

Obsigen · Product · as of 2026-10-22

The Obsigen platform

An AI system that goes beyond an ordinary chat: it works on your company's documents, data and code, picks the models for each task itself and delivers verified results - with full company control over the data.

interface simulation · sample data
591 tools

including 19 central1 and 40 tenant-side1

97 / 982

values read correctly by Obsigen Magnify on a reference document

56 s3 per drawing

reading time of an A0 technical drawing in Obsigen Magnify

1371 reports

from measurements of model quality and cost

Work modes: Normal, Genius, Titan

One assistant, three levels of depth. The modes differ in model strength and in the scope of verification; you choose the mode for every task, and Titan is the default.

modeusewhat it adds
Normalquestions, short tasks, work on files-
Geniusharder reasoning, long documentsbasic GRID orchestration
Titan (default)reports, analyses, code with testsfull GRID orchestration, plan preview

Titan in our test battery: 18/204 - one point below the strongest configuration (19/204), at ≈ 10×4 lower model cost. In Normal mode, the router sends harder questions to a stronger model.

Memory and company glossary

Three levels of memory: the current task, the compacted session, and long-term memory with semantic search.

  • Company glossary: abbreviations, names of systems and departments (e.g., WZ for goods issue notes, MPK for cost centers) - learned from what the user says.
  • User profile: role and preferred answer style.
  • Full control: Obsigen remembers only what the user says, never content from tools. Every saved item shows up as “Remembered · Undo”.
  • Only in your tenant: memory, the company glossary and the document index (RAG) are stored only in your tenant, on infrastructure dedicated to your company.

Team and access

Conversations are private by default. An administrator can turn on team visibility and share chats within the team; invitations have an expiry date. Devices running Obsigen Forge and Obsigen Runtime are listed in the account panel - access for each of them can be revoked at any time.

Tools

59 tools1: 19 central1 and 40 tenant-side1. Where they run: see the architecture description.

files and code
read, edit and search files · run commands · Python / Node / PHP / Lua sandbox
spreadsheets and documents
spreadsheets and pivot tables · document analysis · Word, PowerPoint, charts and diagrams · format conversion · Design
web and public sources
web search and research with cited sources · reading web pages · industry sources: KRS (company register), ISAP (legal acts), NBP exchange rates
images and drawings
product photos (background removal on the tenant server) · image editing · Obsigen Magnify (documents and technical drawings)
memory and communication
long-term memory · clarifying questions · project notes
calculations
calculator · unit conversion · dates and time

Model orchestration

Obsigen picks the model for each task itself. The mode sets the budget; orchestration decides how to use it.

How Obsigen picks the model for a task - and for every subtask A small model classifies the task; a table filled by measurements picks the model; the model does the work and, if the task is larger, splits it into subtasks - each goes through the same cycle, nested too. When the table adds extras - as in Titan mode - a second model reviews the answer and critics check sources and completeness; only what is contradictory is corrected, in a loop. Simple questions are answered straight away by a single model. The result carries a chip: which model, cost, duration. the same cycle for the task and every subtask task question, file,photo classification Obsigen GRID model choice Obsigen GRID execution Obsigen GRID reviewer Obsigen GRID critics sources ·completeness result chip: who · cost ·how long correction: only what is contradictory How Obsigen picks the model for a task - and for every subtask A small model classifies the task; a table filled by measurements picks the model; the model does the work and, if the task is larger, splits it into subtasks - each goes through the same cycle, nested too. When the table adds extras - as in Titan mode - a second model reviews the answer and critics check sources and completeness; only what is contradictory is corrected, in a loop. Simple questions are answered straight away by a single model. The result carries a chip: which model, cost, duration. task question, file,photo classification Obsigen GRID model choice Obsigen GRID execution Obsigen GRID reviewer Obsigen GRID critics sources ·completeness result chip: who · cost ·how long

Models

Open-weight models run by providers in the EEA. A new model joins the lineup, and a change joins the selection table, only after measurement; the full lineup is described in the documentation shared once an NDA is signed.

Selection

A small model classifies the task, and the measurement table picks the model and the extras: a reviewer, a plan preview, escalation. Larger tasks are split into subtasks, and each one goes through the same cycle.

Verification

  1. premise check - whether the question assumes something false;
  2. comparison of the answer with the sources; where they differ, a quote from the source;
  3. correction of the faulty passage only, without rewriting the whole answer;
  4. completeness check.

Where GRID is heading

Model selection is the first layer of GRID. We are not building the next layer as a conversation between models: we are developing a shared, verifiable state of knowledge about the task, to which models contribute hypotheses, artifacts and test results - by reference, without passing content. Every conclusion has provenance, and verification decides what is true. We will publish the details together with the test results.

Obsigen GRID and system cardMeasurement methodology

Analysis

Complex research questions in a single task: splitting into threads, fact verification, a report written section by section, and a review before delivery.

For engineers
  1. scoping - what to look for and where
  2. plan - splitting into threads
  3. parallel research threads, each with its own sources
  4. fact verification - only facts with a source remain
  5. gap analysis - what was not found
  6. drafting the report with references
  7. final review

Obsigen Magnify

Deep Reading mode for technical documents: specifications, bills of quantities and CAD drawings read in a single pass, showing where the values read come from.

  • illegible fragments are marked;
  • an audit overlay that marks where each value comes from;
  • reference document: 97 / 98 values read correctly2.

Product photos

Background removal, shadow and a full set of formats - including Amazon and Shopify, a 1000 × 1000 square and a 9:16 story - from a single request. Output: PNG, layered PSD and ZIP. The background is computed on the tenant server; the product's pixels come from the original photo.

fast preset ≈ 2 s5 · quality preset ≈ 5 s5

Image editing in testing

A new background, headline or composition - with the product preserved.

Design

B2B materials from your data:

  • A4 report with charts and sources · proposal and one-pager · data-driven presentation · instructions with photos · dashboard;
  • visual check by the Obsigen Magnify module, plus verification;
  • PDF and editable PPTX; every chart with its data source stated.

Usage

The usage panel shows the model, tokens and time of every task - without the content of conversations.

Demo · 45 min

See Obsigen on your own example

We show Obsigen live, online, on a document you bring - with no confidential data. Then a quote and deployment on dedicated infrastructure.

Book a demo Let's talk

  1. step 1Demo · 45 minonline, on your example
  2. step 2A one-page quotewithin two business days
  3. step 3Deployment and team starta dedicated tenant and team accounts

Footnotes

  1. build-time counter · 2026-09-28 · built-in tools, as of the build date
  2. Magnify report 2026-06-25 · reading quality check on a reference document
  3. Magnify report 2026-06 · reading time of an A0 drawing
  4. work-mode measurement 2026-08-31 and 2026-09-02 · Titan mode · GRID orchestration, open-weight models
  5. product photo measurement 2026-09-08 §7 · product photo: background removal, fast preset; on the tenant server