Search API V1 — explicit web search for LLMs

Most “LLM with web access” solutions have the same problems:
— fake browsing (“I checked online…”)
— unpredictable behavior
— no control over search
— mixed search + generation
— hard to debug

This project solves it by doing the opposite.

Search is not implicit.
It is explicit.

The model:
— does NOT browse the web
— does NOT verify facts on its own
— does NOT pretend it has internet access

Search is triggered only by a marker:

???

One message → one search → deterministic pipeline.

rewrite → search → rank → extract → (optional pack) → answer

No loops.
No agents.
No hidden behavior.

Everything is:
— controllable
— reproducible
— debuggable

Works with any LLM via OpenAI-compatible API.
Built to integrate with Text Generation WebUI, but runs as a separate service.

This is not “AI magic”.
It’s a clean engineering approach to web search for LLMs.

GitHub:
https://github.com/mistbyte-ai/websearch-mistbyte

If you find this useful, you can support development:
https://web.tribute.tg/d/Ih8

Time Injector — fixing LLM time awareness

Local LLMs don’t know the current date or time.

This leads to annoying behavior:
— “It is still 2024”
— refuses to discuss recent events
— confuses today / yesterday / tomorrow
— wrong greetings (“Good morning” at night)

This extension fixes it by adding a small machine-readable time block to every prompt.

Injected into the model:

CURRENT_UTC_DATETIME
CURRENT_LOCAL_DATETIME
UTC_OFFSET_MINUTES
LOCAL_TIME_OF_DAY

That’s it.

No hacks.
No guessing.
Just a reliable time anchor.

Works in all modes (chat / instruct / notebook).
No UI noise.

Small detail — but it significantly improves consistency.

GitHub:
https://github.com/mistbyte-ai/time-injector-mistbyte

If you find this useful, you can support development:
https://web.tribute.tg/d/Ih8

MistByte AI — local-first tools for LLMs

This channel is about practical tools for working with local language models.

Focus:
— web search integration
— time/context consistency
— long-context workflows
— infrastructure around LLMs

No “AI magic”.
No fake capabilities.

Only explicit, controllable systems.

Current projects:
— Search API (explicit web search for LLMs)
— Time Injector (fixes time awareness)

More tools are coming:
— long-context browser integration
— repo-to-LLM workflows

GitHub:
https://github.com/mistbyte-ai

Portfolio:
https://home.vps.3-a.net/

If you find this useful, you can support development:
https://web.tribute.tg/d/Ih8