01 / 03 · REQUEST
Start with the right information.
Ask your question and select the approved documents or a knowledge store. The workflow uses the content intended for your task.
Access permissions and sources are configured in advance.FOR YOUR IT TEAM
How Mikis works with your data.
We tailor search, language models and additional services to your task. The model and supporting services must all fit within the agreed security boundary.
SECURE. CLEAR. PRACTICAL.
From your question to a result you can check.
02 / 03 · PROCESSING
Search and the language model work together.
Search retrieves relevant passages. The language model processes your request with this context. Text recognition and other supporting services are also part of the agreed data flow.
A database module can be added for structured calculations.03 / 03 · REVIEW
Check the sources. Assess the result.
Your team checks statements, sources and the data used. We test typical tasks and access boundaries together before approving the workspace for use.
Even with sources, AI can miss or misinterpret information.Keyword search, semantic search and calculations
Search your documents. Mikis combines matching terms with semantic similarity. An additional step can rank source passages by relevance.
Calculate spreadsheet figures transparently. We can configure an additional database for structured data, allowing specific values to be selected and calculated.
We test the setup against typical questions from your work. Data and integrations in detail.
Models for different tasks
We select language models and supporting services for your tasks. Your agreed setup determines which functions and models your team can use.
Text and analysis
Answer questions, draft text and organise information.
Understand images
Include image content when using a model that supports it.
Read documents
Extract text from suitable documents and scans for further processing.
Search your knowledge
Prepare semantic search and rank relevant sources.
Work with speech
Transcribe audio or read text aloud when these services are configured.
Model overview · As of 26 October 2026
This overview records the position as of 26 October 2026. We confirm which models are available and suitable for your installation during setup.
| Model | Purpose | Provider status |
|---|---|---|
| gpt-oss-120b | Text, analysis, tools | LTS |
| Qwen3.5-0.8B | Small text tasks | Stable |
| Ministral-3-14B-Instruct-2512 | Chat and image understanding | LTS |
| Qwen3.5-122B-A10B-FP8 | Text and images | Stable; deactivated from 28 December 2026 |
| Qwen3.6-35B-A3B-FP8 | Long text and image tasks | Stable |
| Qwen3.8-27B-NVFP4 | Code and tool workflows | Stable |
| GLM-OCR | Extract text from documents and scans | Stable |
| Qwen3-Embedding-8B | Prepare semantic search | LTS |
| Qwen3-VL-Reranker-2B | Reorder search results | Stable |
| whisper-large-v3-turbo | Convert audio to text | LTS |
| cohere-transcribe-03-2026 | Transcripts with speakers and timestamps | Testing phase |
| Qwen3-TTS-12Hz-1.7B-CustomVoice | Read text aloud | Stable |
The model name and processing location are different things.
A model can run in your infrastructure even if it was developed by another company. What matters is the service that actually operates it. A compatible API format does not, by itself, mean data is sent to the model developer.
Open-weight models have available model weights under their respective licences. This says nothing about the developer’s country of origin or whether the entire software is open source.
LET'S GET SPECIFIC
Discuss your task and the right deployment.
Let's discuss your documents, security requirements and the right technical setup.