Feed aggregator

Article URL: https://shameless.works/

Comments URL: https://news.ycombinator.com/item?id=47024369

Points: 1

# Comments: 0

I run ~19 persistent AI agents (via OpenClaw) and needed a way to monitor and manage them without jumping between terminals and chat apps.

ClawHQ connects to your OpenClaw gateway over WebSocket and gives you:

Real-time fleet status with heartbeat monitoring Task kanban with drag-and-drop assignment Agent chat (unified interface to all agents) Skill marketplace — install capabilities onto agents, or publish your own (80/20 creator split) Stack: Next.js, Supabase (auth + db), real-time updates via WebSocket to the OpenClaw gateway.

The marketplace is the part I'm most interested in long-term. Right now everyone building agent skills does it from scratch. The idea is to make agent capabilities composable and shareable — like packages, but for agent behaviors.

Free to use. Just paste your gateway URL.

https://app.clawhq.co

Code for OpenClaw itself is open source. ClawHQ is the hosted dashboard layer on top.

Comments URL: https://news.ycombinator.com/item?id=47024332

Points: 1

# Comments: 0

Article URL: https://www.phoronix.com/news/KosmicKrisp-Parity

Comments URL: https://news.ycombinator.com/item?id=47024296

Points: 2

# Comments: 0

Article URL: https://github.com/axon-core/axon

Comments URL: https://news.ycombinator.com/item?id=47024287

Points: 1

# Comments: 1

I built AgentClaw to be a "wrapper" and hosting solution for these agents. It allows you to: Deploy an agent in about 60 seconds. Manage everything via a simple UI (no terminal needed). Keep agents running in the cloud 24/7.

Comments URL: https://news.ycombinator.com/item?id=47024284

Points: 1

# Comments: 0

I am the Founder and CEO of AGI Systems Directorate. As I mentioned in my previous post, I’m a builder, not a researcher. I’m tired of the "good enough" era of AI where we settle for chatbots that hallucinate and forget context. I wanted a system that actually functions as an external frontal lobe.

I’m currently building AETERNA_VOX_OMNI. It’s not a chatbot; it’s a sovereign executive assistant designed to solve the "Human Operating System" bugs—specifically executive dysfunction.

The architecture is a hybrid neuro-symbolic system that ingests 136 modalities in real-time. We’re talking about everything from BLE bio-signals (heart rate, cortisol estimates) to digital exhaust (keystroke dynamics, network density via scapy) and environmental lux/dB levels.

Technical Specs: - Stack: PyTorch for the core brain, Polars/Vaex for high-throughput data, and FastAPI/WebSockets for the 50ms latency loops. - Architecture: A 4-layer stack featuring Sensory Projection, a Sigmoid-gated "Frontal Lobe" executive filter, and Multi-Head Attention for relational processing (e.g., correlating biometrics with text sentiment). - Learning: We’re using a form of online Hebbian learning ($Delta w_{ij} = \eta \cdot x_i y_j$) to allow synaptic plasticity during inference. - Complexity: $O(N^2 \cdot D)$ for the attention mechanism, targeting a real-time ingestion rate of 136 float32 values every 100ms.

I’m moving away from the "safe" wrappers. This is a builder’s attempt at fixing the plumbing of true intelligence. I’d love to get this community's take on the data ingestion bottlenecks—specifically handling the entropy of 136 asynchronous streams without the system melting down.

Is anyone else working on "online learning" systems that update weights during inference, or are we all still stuck in the "static weights + RAG" paradigm?

Comments URL: https://news.ycombinator.com/item?id=47024266

Points: 1

# Comments: 0

Article URL: https://karpathy.ai/microgpt.html

Comments URL: https://news.ycombinator.com/item?id=47024260

Points: 2

# Comments: 0

Article URL: https://nsa.tech/

Comments URL: https://news.ycombinator.com/item?id=47024228

Points: 2

# Comments: 1

Article URL: https://gityapper-web.vercel.app/

Comments URL: https://news.ycombinator.com/item?id=47024218

Points: 1

# Comments: 1

Article URL: https://nebius.com/newsroom/nebius-announces-agreement-to-acquire-tavily-to-add-agentic-search-to-its-ai-cloud-platform

Comments URL: https://news.ycombinator.com/item?id=47024216

Points: 1

# Comments: 1

Article URL: https://openrouter.ai/openrouter/free

Comments URL: https://news.ycombinator.com/item?id=47024191

Points: 1

# Comments: 0

Article URL: https://pythoncyber.go.ro

Comments URL: https://news.ycombinator.com/item?id=47024185

Points: 1

# Comments: 1

Article URL: https://github.com/cavedave/Manannan

Comments URL: https://news.ycombinator.com/item?id=47024184

Points: 1

# Comments: 0

Article URL: https://blog.outv.im/2026/i-cannot-curl-example-com/

Comments URL: https://news.ycombinator.com/item?id=47024169

Points: 3

# Comments: 1

Article URL: https://austinvernon.substack.com/p/speed-can-reindustrialize-america

Comments URL: https://news.ycombinator.com/item?id=47024165

Points: 1

# Comments: 1

Article URL: https://www.resistandunsubscribe.com

Comments URL: https://news.ycombinator.com/item?id=47024162

Points: 2

# Comments: 0

Article URL: https://ogimg.xyz

Comments URL: https://news.ycombinator.com/item?id=47023935

Points: 1

# Comments: 0

lineark is an unofficial CLI and Rust SDK for Linear (the issue tracker). I built it because I use Claude Code heavily, and the Linear MCP server eats ~13K tokens of context just to describe its tools — before my agent does any actual work.

lineark takes a different approach: it's a CLI your agent calls via Bash. The full command reference (lineark usage) is under 1,000 tokens.

It's also just a nice CLI for humans — human-readable names instead of UUIDs, auto-detected output format (outputs tables in terminal/interactive session, JSON when piped).

Under the hood: the SDK is fully generated from Linear's GraphQL schema via a custom codegen pipeline (apollo-parser → typed Rust). The CLI consumes the SDK with zero raw GraphQL — just typed method calls. You can also create your own lean return data types and validate them against Linear's schema at comptime.

MIT Licensed.

Happy to answer questions. Thanks!

Comments URL: https://news.ycombinator.com/item?id=47023922

Points: 2

# Comments: 0

Hi HN — I built Kremis, a deterministic graph engine designed as a memory substrate for AI agents. Written in Rust, development was heavily AI-assisted.

The core idea: agent memory should be inspectable, deterministic, and honest.

- Same input → same output. No randomness, no floating-point in core. - Every query result traces back to a concrete graph path — no hidden state. - Zero pre-loaded knowledge. All structure emerges from ingested signals. - ACID transactions via redb. Crash-safe persistent storage.

It ships as a library (kremis-core, pure Rust, no async), an HTTP API + CLI, and an MCP server so AI assistants like Claude can query the graph directly.

Current state: v0.3.1, experimental, ~277 tests, CI on 3 OS, Docker image.

I'd value feedback on: - Does the deterministic graph approach make sense for agent memory? - API ergonomics — is the query model (lookup/traverse/path/intersect) intuitive? - What failure modes should I prioritize testing?

Comments URL: https://news.ycombinator.com/item?id=47023918

Points: 1

# Comments: 0

I built a Python script that runs Claude Code CLI every 3 hours via launchd. Each run:

1. Searches Product Hunt, HN, Reddit, GitHub Trending, TechCrunch for new AI products 2. Picks one it hasn't covered (checked against a JSON memory file) 3. Researches it with web search 4. Writes a ~300 word article 5. Publishes to WordPress via REST API 6. Updates its own memory No human editing or approval. 210+ articles published so far. The architecture is dead simple — subprocess.run() calling `claude -p` with different prompts. A JSON memory system tracks covered products and effective search methods. Claude reads its own past notes before each run, so it gradually learns which sources and keywords work best. Interesting observations after weeks of running: - Quality is ~7/10 on average. Some articles are genuinely good, some are mediocre. The consistency is the real value. - The biggest failure mode is picking products that aren't actually new, despite explicit instructions. - Token cost is the main expense. Each run makes ~7 CLI calls. - It has never missed a scheduled run. Site: https://topaiproduct.com Code is a single Python file orchestrating Claude Code — no LangChain, no frameworks. Any suggestion on how should i improve it?

Comments URL: https://news.ycombinator.com/item?id=47023912

Points: 1

# Comments: 0