Feed aggregator

Elvebredd for iOS

Hacker News - Fri, 07/10/2026 - 2:01am
Categories: Hacker News

Promising inference software, written in Zig

Hacker News - Fri, 07/10/2026 - 1:56am

Article URL: https://zml.ai/posts/llmd/

Comments URL: https://news.ycombinator.com/item?id=48856255

Points: 1

# Comments: 0

Categories: Hacker News

TinyToT – Tree of Thoughts Inference Server

Hacker News - Fri, 07/10/2026 - 1:53am

Article URL: https://github.com/guilt/TinyToT

Comments URL: https://news.ycombinator.com/item?id=48856229

Points: 1

# Comments: 1

Categories: Hacker News

Neuralwatt Doubling the Price

Hacker News - Fri, 07/10/2026 - 1:34am
Categories: Hacker News

Show HN: Tempest – an open-source, token-efficient agent orchestration platform

Hacker News - Fri, 07/10/2026 - 1:14am

I've been building Tempest, an open-source token-efficient agent orchestration platform for AI coding agents.

Most orchestration tools focus on running more agents. Tempest focuses on making each agent use fewer tokens.

Tempest runs on top of your existing coding agents... Claude Code, Gemini CLI, GitHub Copilot CLI, Opencode, Goose, Cline, Cursor Agent, and others... so there's no separate AI subscription or usage billing from Tempest.

Instead of giving every agent the same repository context, Tempest builds a knowledge graph of the codebase and retrieves only the code that's relevant to the current task. In our testing, this reduced token usage by 64% compared with native CLI coding agents without increasing token usage.

Each agent runs in its own Git worktree inside an isolated process, so multiple agents can work on the same repository without interfering with one another. To make that work across platforms, we built Hephaestus, an open-source SDK for process isolation. It uses the native sandboxing mechanisms on each operating system... Seatbelt on macOS, Bubblewrap on Linux, and a custom implementation we built for Windows.

The goal isn't to increase context windows. It's to improve repository understanding. Better understanding lets agents spend fewer tokens searching through code and more tokens solving problems.

If you've used tools like Conductor, I'd love to hear how you've approached parallel execution, repository understanding, or token efficiency. I'm especially interested in feedback on the architecture and the overall developer experience.

Comments URL: https://news.ycombinator.com/item?id=48855970

Points: 1

# Comments: 0

Categories: Hacker News

Semantic Fingerprinting Made Easy

Hacker News - Fri, 07/10/2026 - 1:13am

Article URL: https://github.com/thorwhalen/guise

Comments URL: https://news.ycombinator.com/item?id=48855966

Points: 1

# Comments: 0

Categories: Hacker News

Pages