Hacker News

Show HN: oMLX – Native Mac inference server that persists KV cache to SSD

Hacker News - Thu, 02/19/2026 - 5:09pm

I built an open-source LLM inference server optimized for Apple Silicon. The main motivation was coding agents - tools like Claude Code send requests where the context prefix keeps shifting, invalidating KV cache. A few turns later the agent circles back, and your Mac has to re-prefill the entire context from scratch.

oMLX solves this with paged SSD caching. Every KV cache block is persisted to disk. When a previous prefix returns, it's restored instantly instead of being recomputed. This makes long coding sessions significantly faster.

It also supports continuous batching for concurrent requests, multi-model serving (LLM + embedding + reranker) with LRU eviction, block-level KV cache with prefix sharing and copy-on-write, OpenAI and Anthropic compatible APIs, and tool calling. Ships as a signed macOS menubar app with a web dashboard.

GitHub: https://github.com/jundot/omlx

Comments URL: https://news.ycombinator.com/item?id=47080277

Points: 1

# Comments: 0

Categories: Hacker News

Show HN:PolyMCP Python Tools with Autonomous Agents

Hacker News - Thu, 02/19/2026 - 5:07pm

PolyMCP is a framework that exposes functions (in Python or TypeScript) as MCP tools, serves them through standardized MCP servers, and orchestrates them with autonomous agents capable of planning and executing complex workflows.

With PolyMCP, you can: 1. Expose existing functions (in Python or TypeScript) as MCP tools without rewriting the code. 2. Use the PolyClaw agent to dynamically orchestrate tools, break down complex tasks, and adapt to failures, all running in isolated environments via Docker.

PolyMCP is ideal for enterprise automation, DevOps, data pipelines, and internal tool orchestration.

GitHub Repo: https://github.com/poly-mcp/PolyMCP

Comments URL: https://news.ycombinator.com/item?id=47080258

Points: 1

# Comments: 0

Categories: Hacker News

The Observatory of Anonymity

Hacker News - Thu, 02/19/2026 - 5:06pm

Article URL: https://www.ooa.world/

Comments URL: https://news.ycombinator.com/item?id=47080252

Points: 2

# Comments: 0

Categories: Hacker News

Human Flatus Atlas

Hacker News - Thu, 02/19/2026 - 5:05pm

Article URL: https://www.flatus.info

Comments URL: https://news.ycombinator.com/item?id=47080241

Points: 1

# Comments: 0

Categories: Hacker News

Show HN: CursorLens – Open-source screen recorder/editor for product demos

Hacker News - Thu, 02/19/2026 - 4:12pm

Hi HN — I built CursorLens, a free and open-source screen recorder/editor for product demos and walkthroughs.

It started as a fork of OpenScreen, but I reworked a lot of the capture/editing pipeline for a more native macOS workflow.

Current features:

full-screen or window recording native cursor hide/show behavior on macOS camera overlay + microphone recording timeline editing (trim/crop/zoom/cursor effects/annotations) subtitle generation + rough-cut workflow multi-aspect export (16:9, 9:16, 1:1) and batch export audio export controls (gain, loudness normalization, limiter) It’s still in beta, so stability can vary by machine.

Repo: https://github.com/blueberrycongee/CursorLens Releases: https://github.com/blueberrycongee/CursorLens/releases

I’d really appreciate feedback on:

recording reliability across setups editing UX (especially rough cut/subtitles) export speed/quality tradeoffs

Comments URL: https://news.ycombinator.com/item?id=47079466

Points: 1

# Comments: 0

Categories: Hacker News

Ask HN: What makes AI agent runtime logs defensible under adversarial audit?

Hacker News - Thu, 02/19/2026 - 4:12pm

Modern AI agents can execute tools, write to databases, and trigger irreversible actions.

Most teams rely on traditional logging (OpenTelemetry, SIEM, DB audit logs). But under adversarial conditions (audit, litigation, incident response), those logs depend on platform trust and cannot typically be verified independently of the system that produced them.

I’m exploring whether agent runtime evidence should be:

-deterministically canonicalized

-hash-chained

-signed

-optionally externally timestamped

-verifiable offline

The goal isn’t observability. It’s defensibility.

Open questions:

1.Is RFC 3161-style timestamping sufficient to deter practical backdating, or is some form of multi-witness anchoring necessary?

2.In real distributed agent systems, where does replayability break down?

3.At what scale or risk threshold does this move from overengineering to necessary?

Not looking for blockchain/ledger answers — more interested in models that integrate with existing infrastructure.

Trying to understand whether this addresses a real integrity gap in production systems.

Comments URL: https://news.ycombinator.com/item?id=47079460

Points: 1

# Comments: 0

Categories: Hacker News

Show HN: Auto-scrolling for WhatsApp group chats

Hacker News - Thu, 02/19/2026 - 4:09pm

I got tired of scrolling through endless WhatsApp group chats, so I built a browser extension that does the scrolling for me. You can tag contacts as "important" to scroll slower when their messages are in view. You can also hide messages from specific users (click-to-unhide like Slack has).

Comments URL: https://news.ycombinator.com/item?id=47079430

Points: 1

# Comments: 0

Categories: Hacker News

Pages