llm-news
LiveA live tech news feed, curated and ranked by AI.
Aggregates and ranks tech and AI news from 70+ sources, refreshed hourly. Articles are summarized and rated for relevance; thematic clusters surface evolving stories.
LLM Works
AI agents, from research to production.
LLM Works builds AI agents on its own open-source substrate, and turns the newest research into production-grade systems that power existing pipelines or create new ones from scratch.
Get something built
Agents built from cutting-edge research, running in your production pipeline.
New agent methods appear every month, but very few reach real-world production. We distill the best of them into our platform, hardened and ready to deploy in your pipeline, with you.
Built on the open source substrate. What we deploy stays yours.
See what it runs on
The next-generation open-source architecture for AI agents.
Multi-layer library stack covering all aspects of semi-autonomous agents implementation: typed contracts around the model, verification loops, tracing, memory and learning.
Reasoning stays in the model. Guarantees live outside it.
Agents
Highly refined & purpose-built loops.
Our production agents are custom-built and optimized to solve a specific problem. We deploy state-of-the-art loop techniques and combine them with suitable existing methods from science and technology, such as k-means clustering and knowledge graphs.
llm-xray
LiveDeep research powered by AI.
Decision-grade research output: calibrated verdicts through self-refining loops, with a plan-as-protagonist architecture that surfaces what was checked, what changed the verdict, and what's still open.
Platform
An open-source substrate for the engineering layer around production LLM agents
- Five independent modules, each usable on its own.
- Clean on-LLM / off-LLM boundary via verb pattern.
- Response guards enforce contracts on model output.
- Model-agnostic inference across providers.
- Cost transparency and per-call tracking.
- Rails for composable and automated agent loops.
- Clean code, strict typing, ~9k tests, 1:1 test-to-source ratio.
Every agent runs on a modular architecture enabling a clean on-LLM / off-LLM boundary.
The engineering layer around the model turns model calls into production agents. The platform builds it once, so every agent inherits the capabilities.
Monolithic design
Platforms that ship as one blob are hard to compose, extend, or swap parts of. Adoption is all-or-nothing and hard to walk back.
Lack of abstractions
Without higher-level abstractions, business logic becomes complex. Every model call turns into request formatting, response parsing, and retry handling.
Maintenance overhead
Every dependency, upgrade, and breaking change costs engineering hours. The maintenance bill grows with the codebase.
Story
AI is going through an engineering shift seen in several industries before