Chat

Thinking Incognito

History

No conversations yet

Role

No roles yet — create one in the Role tab

Start a conversation…

Translate

From

Mode

Source text

Translation

Result will appear here…

WebAsk

Question

URL (optional)

Top N

Result will appear here…

WebCrawl

URL

Mode

Filter

URL prefix (optional — defaults to start URL)

Output path (optional — default: ~/.vyrii/crawl/)

Depth

Max pages

Ask LLM summary

Task / goal

Columns (YAML / XPath)

Result will appear here…

WebAnalys

Query

Results

Result will appear here…

DeepAgent MD

Task / Topic

Reference URL (optional)

Sections

Web search RAG Team

Result will appear here…

Interview

Task description

Number of questions

Scan

Path

Query (optional)

Chunk

Summary

Target

Rounds

Extensions

Result will appear here…

WebIndex

URL

Project name

Output path (optional)

Depth

Max pages

Result will appear here…

Obfuscate / Deobfuscate

Obfuscate

Text

Dictionary

Force

Result will appear here…

Deobfuscate

Text

Dictionary

Force

Result will appear here…

RAG

Project

Top K

Query

Results

Result will appear here…

Settings

Connection

Ollama URL

Backend

Active profile

Reserve mode

Till end of response By timer s

Timeouts

Request timeout (s)

Worker timeout (s)

Defaults

Default model

Language

Theme

External Tools

Authentication

Username

New password

Ollama daemon

KV cache type (OLLAMA_KV_CACHE_TYPE)

Flash attention (OLLAMA_FLASH_ATTENTION=1)

Keep alive (OLLAMA_KEEP_ALIVE)

Max loaded models (OLLAMA_MAX_LOADED_MODELS)

Host (OLLAMA_HOST)

Vulkan backend (OLLAMA_VULKAN=1)

Enable integrated GPU (OLLAMA_IGPU_ENABLE=1)

Diagnostics

Measure the maximum context size this model/host/timeout combination can handle before failing.

Search mode

Full mode (wait for entire response)

Start (tokens)

End (tokens, bin only)

Step (tokens)

Saved report Confirm delete

AutoCut Context

Non-destructive sliding-window projection of the chat history (fixed head + sliding tail + optional mid-selection) applied before each LLM call, once KV-cache stops helping.

Enable AutoCut Context

First (tokens)

Last (tokens)

Mid limit (tokens)

Mid algorithm

System control

Restart command

Saved automatically. Default: vyrii --flask

Confirm dangerous action

Files

Upload

Loading…

Select a file or folder

Glossary

Index a folder

Folder (relative to ~/.vyrii)

Project

Chunk (tokens)

Overlap (tokens)

Browse

Glossary (folder / project)

Search terms

Profiles

Saved

Name

Comment

Workers

Колектив

Profile

Combine

Context

Query

Result will appear here…

Scheduler

Tasks

Add new task

Name

Schedule type

Command

Time HH:MM

Day of week

Interval value

Task ID (first 8 chars)

View task logs

Projects

Project registry — name to local path. Used by simargl and svitovyd tabs.

Add project

Name

Path

Description (optional)

simargl

Task-to-code retrieval — index a project, then search by task description.

You have a task. You do not know which files to edit. simargl reads git history and finds the most likely files.

Project

Reads all git commits in this project. Builds a semantic index. Do this once before searching.

Result will appear here…

svitovyd

Project map — index code structure, then find/trace/deps/sym/keywords.

Scan your code and build a map of all functions, classes, and their links. Then explore the structure.

Project

Scan the project folder. Find all function and class definitions. Save links between them. Result: .svitovyd/map.txt.

Depth 2 — definitions and calls. 3 — also variables and parameters. Use 2 for most projects.

          Result will appear here…
        

Prompts

Prompt library — save and search by name, model, or area.

Add prompt

Name

Description

Model

Area

Prompt text

Role

Named system prompts chosen once at the start of a new chat (via the Role button in Chat) — stay invisible for the rest of it. Distinct from Prompts, which are inserted manually.

Add role

Name

Reserved size (tokens)

System prompt

~0 tokens