iCreatAI News
Back to main site
Skip to content
#Open source
Nov 6Fri
  1. Hacker News AI58

    Show HN: Shotoku – Local-first action governance and budget caps for AI agents

    Shotoku is an open-source, local-first tool that enforces budgets, approval gates and a tamper-evident audit trail for AI agent spending before payments occur, using policy rules in a YAML file and a TypeScript authorize function. It holds no funds or credentials, stores decisions in a local JSONL file, and offers MCP wrapping, signed receipts, and Telegram-based approvals. It is in beta and runs end to end against PayPal's sandbox.

Nov 5Thu
  1. Hacker News AI34

    Show HN: Northmeadow – encrypted shared memory for local AI agents and people

    Northmeadow 0.1.0-alpha.1 is a public alpha of encrypted shared memory for local AI agents and people, distributed via a curl-pipe install script or a manual download with checksum and ED25519 signature verification. The unfinished test software is licensed until 31 January 2027 and states it does not comply with the EU Cyber Resilience Act (Regulation (EU) 2024/2847). The alpha is initialised with fixed values including IC_ALPHA_PLAN_CODE public_alpha, 50 clusters and 50 agents, and IC_ALPHA_AUDIT_EXPORT true.

  2. Hacker News AI40

    Show HN: MaidenRun – Deploy an AI agent to your AWS and collect tester feedback

    MaidenRun is an open-source tool that deploys an AI agent from its GitHub repo to your own AWS account, provides a shareable link, and collects tester ratings and comments through a built-in feedback form and dashboard. It supports redeploying to the same address to compare feedback across versions, stores secrets in AWS Parameter Store, and offers teardown that deletes all created resources. It requires Docker and an AWS account, with the agent's UI on port 8001 and ports 8002 and 8003 reserved for the feedback form and dashboard.

  3. Hacker News AI58

    Show HN: Revline – An open-source AI-native video editor

    Revline is an AI video editor that lets users upload footage and describe the desired edit in conversation, with the agent handling cuts, captions, motion and sound on an editable timeline. It targets podcast clips, social shorts, motion design and talking-head videos, with features including animated captions, background removal, masking and 9:16/1:1/16:9 export. It costs $39 USD per month after a 3-day free trial, and the post says no credit card is required to start.

  4. Hacker News AI66

    AgentFleet – Cross-device AI agent fleet: 303 tasks/s, 0 lost on failover

    AgentFleet is a self-hosted orchestration layer that coordinates already-installed AI coding agents (Claude Code, Codex, Cursor, Aider, Gemini, Qwen, Grok and others) across machines, with a live neural graph dashboard and a tamper-evident shared memory ledger using a hash chain, Merkle tree and Ed25519 signatures. It reports 303 tasks/s scheduler throughput on 200 tasks (4 devices × 2 agents), 377 tasks/s on 1000 tasks, ~1s disconnect detection, and zero tasks lost on failover, with an OpenAI-compatible routing gateway, Excel-based batch deployment, and an A2A endpoint. The project states that only Grok has a stable local CLI direct adapter, while other platforms have discovery and dispatch scaffolding pending per-platform verification.

  5. Hacker News AI22

    Abra, an AI creator crew built on Meta's Muse models

    Abra is a hackathon project (Meta Global AI Developer Hackathon, Indie Dev Track, Oct 26 – Nov 4, 2026) that builds an AI creator studio on the Meta Model API and Claude, with three crew characters: Nico for comment replies, Sage for turning a dropped video into a post, and Koa for earnings and legal tracking. It calls five hosted Meta models — Muse Wonder, Muse Spark, Muse Image, Muse Voice Transcribe and SAM 3.1 — plus Claude Opus for in-voice replies and frame vision, and runs only on macOS with Python 3.12 and ffmpeg. The author notes that the comment-reply loop and first dashboard snapshot predate the event, that Muse Image's overlay path is not yet tested end to end, and that several features (camera, Greenscreen, Create templates, voice) are switched off for the submission.

  6. Hacker News AI38

    WebsiteBench: Can AI Agents Rebuild Websites Through Browser Exploration?

    WebsiteBench is a benchmark in which agents explore a production-level website through a browser and must reverse-engineer a full-stack version matching its appearance and behavior as closely as possible. The repository provides quickstart instructions using uv, Docker, and the harbor run command to evaluate single tasks from local-tasks/ or all tasks, with results written to jobs/. Code is licensed under Apache-2.0, while task data and reference content retain their applicable licenses.

  7. Hacker News AI58

    Show HN: Ergon - build AI agents in code or on a canvas that generates Python

    Ergon is a monorepo toolkit for building AI agent harnesses, combining a Python SDK engine with a drag-and-drop visual Studio that generates Python code against it. The SDK covers tool-calling loops, control flow, a unified router for roughly 15 providers, RAG, MCP clients, browser control, OpenTelemetry traces and an evaluation harness, while the Studio adds a canvas, inspector, live preview and codegen. A bootstrap script sets up both halves, and harnesses can be served as REST + SSE endpoints or deployed via Docker with Postgres/pgvector and Redis.

  8. Hacker News AI34

    Smart-contract risk reports that AI agents buy per call over x402

    Magneta Finance runs a production service that sells deterministic on-chain smart-contract risk reports to AI agents for $0.01 per call in USDC via x402, exposed as an HTTP route (GET /v1/risk/:chainId/:address) and a stateless MCP tool (contract_risk_summary), both backed by the same analyzeContract function using public data only (RPC, Blockscout, Sourcify, a local aderyn binary) with no LLM. Reports cover proxy detection, ERC-20 probes, opcode and selector heuristics, verification status and aderyn findings, scored 0-100 into low/medium/high/critical levels, with a 24 s deadline and 10-minute in-memory cache; only 200 responses are settled, and a free bytecode-only preview is rate-limited to 30 requests per minute per IP. It is live at https://x402.magneta.finance on Base mainnet with the Coinbase CDP facilitator, and ships a client package, agent skill, ElizaOS plugin and ERC-8004 registration file.

  9. Hacker News AI34

    Show HN: Telegram AI assistant with projects, images and background jobs

    A private, MIT-licensed TypeScript Telegram assistant offering text and voice conversations, photo analysis and editing, image generation, web search with sources, document questions, projects, named conversations, explicit saved memory, usage controls, streaming replies and durable background jobs. It requires Node.js 26.13+, npm 11, a Telegram bot token and an OpenAI project API key, with FFmpeg/libopus needed only for outgoing voice messages. State is stored locally in SQLite, and the README details commands, limits, data handling and a unit-test suite with a 100% coverage gate.

  10. Hugging Face Blog71

    A Closer Look at Phonon TTS. Launching High-Quality On-Device Speech That Tops the Benchmarks.

    Gradium details Phonon, a small on-device text-to-speech model released two weeks ago that needs no server or network, with Phonon 31m (English) under Apache 2.0 and Phonon 40m covering English, French, German, Spanish and Portuguese under a non-commercial license. The models, about 65 MB and 75 MB, offer over 100 built-in voices plus voice cloning and voice design, and on Hugging Face's Open TTS Leaderboard Phonon ranks first in German and Spanish on intelligibility and tops the CPU streaming benchmark on time-to-first audio. The post also covers the ptts inference framework, Transformers integration from v5.21, launch demos, and more than 200,000 downloads in the first two weeks.

    Why it matters: It offers a detailed look at how a small on-device TTS model was built, benchmarked and integrated into developer tooling.

  11. Hacker News AI34

    Faaah (Filesystem as an AI Handler) – Agent Subscription as OpenAI API via Files

    FAAAH is a dependency-free, 308-line Python proxy that turns an OpenAI-compatible endpoint into a plain-text file protocol: it writes incoming requests to prompt-<id>.txt files in a queue directory, and an existing AI coding agent (e.g. Claude Code, opencode) reads them and writes responses to response-<id>.txt files, which FAAAH packages back into OpenAI-compatible JSON. It supports the OpenAI tools parameter with multi-turn tool loops, optional local embeddings via a uv extra, and a human-in-the-loop fallback where you can type the answer yourself; streaming is not supported. The author frames it as reusing an AI agent subscription for weekend side projects and notes caution about massive datasets and provider ToS.

  12. MarkTechPost39

    Google AI Research Releases Agile Deliberation Framework for Gray-Area Image Classifiers

    Google Research released Agile Deliberation, a human-in-the-loop framework that helps users turn vague concepts like 'healthy food' into structured written definitions that also serve as prompts for a Gemini 2.5 Flash VLM classifier. It finds borderline images by meaning rather than classifier confidence, using borderline queries, CLIP/FAISS retrieval, dictionary-learning clusters and a UCB bandit, and averaged 10.5 F1 points above zero-shot Gemini 2.5 Flash across 2 concepts and 18 live sessions. The code is Apache 2.0 on GitHub with a demo notebook, and the work appears in a CVPR 2026 paper.

  13. Hacker News AI58

    AI isn't slop. It's just the generic average. And not you

    Nahum Litvin introduces voiceprint, a Claude Code plugin that builds a personal writing baseline from pre-AI texts (docs, email, WhatsApp, Slack, old posts) and scores every PR, commit, Slack message or email the agent sends against it, flagging AI-tell words, em dashes and long sentences. The post contrasts an agent-written PR description scoring 5 against his own rewrite scoring 0, and describes an optional Haiku hook that judges whether a draft sounds like him. voiceprint is open source, requires Python 3.10+, and stores data in ~/.voiceprint.

  14. Hacker News AI36

    Rack Community: a self-hosted AI lab whose job runner won't overload the box

    Rack Community is a self-hosted, single-box AI lab with live host gauges, a durable job runner that admits jobs only when their declared memory peaks fit under a limit, recipe-based job templates, a human-approved memory control plane (Somnus), and a team room (TalkBack) that AI agents can join. It ships no model or keys, works with Ollama or any OpenAI-compatible endpoint, starts in dry-run, and binds to 127.0.0.1 with token-checked writes. Version 1.0.0 is the first public release under MIT.

  15. Hacker News AI58

    Bansat – Make your AI agent write the blueprint before the code

    Bansat is a blueprint-first framework and Agent Skill for Claude that has an AI agent write a full plan covering product, architecture, security, privacy law, accessibility and UI states before any production code, then wait for approval. It defines four gates (G1–G4) blocking code, real personal data and launch, plus 17 blocking launch traps and a retrofit track for apps already live. The framework is a single Markdown file (15 parts, 4 appendices, about 2,800 lines) installable via git clone into ~/.claude/skills/bansat, with text and templates under CC BY 4.0 and code samples under MIT.

  16. vLLM Blog72

    Introducing Kinference: An Open Source Inference Control Plane for vLLM and Kubernetes

    Buoyant announced Kinference, an Apache v2 open source control plane for building inference platforms on vLLM and Kubernetes, offering token rate limiting, tenant usage and cost tracking, KV-cache-aware load balancing, and a YAML-based inference policy engine. The post reports benchmarks in which Kinference cut median time-to-first-token by 95% and median end-to-end latency by 45% versus standard Kubernetes load balancing on one workload, and served 7.4× the goodput on a shared-prefix workload and about 3.5× on multiturn. It also covers per-tenant inference metrics and Helm-based installation steps.

    Why it matters: It gives a concrete look at cache-aware routing and policy-driven model selection for self-hosted vLLM deployments, backed by published benchmarks.

  17. Hacker News AI74

    Show HN: BrowserForge – AI agents use your real browser, you approve the clicks

    BrowserForge is a local-first Firefox and Chrome extension plus MCP server that lets AI agents like Claude Code and OpenAI Codex work in your real browser session, with human approval required for consequential actions such as submits, sends, payments, passwords and uploads. It captures observed traffic and turns it into OpenAPI 3.1 specs and Python/TypeScript/JavaScript/PHP clients, and records agent actions as replayable flows exportable to Playwright, Selenium, Chrome DevTools Recorder or JSON. Everything runs on 127.0.0.1 with token-authenticated transports, secret redaction and page content treated as untrusted data; version 0.2.1 is published on the Chrome Web Store and Firefox Add-ons under the MIT license.

  18. MarkTechPost76

    Ricursive AI Releases Ripple-65B Preview: A 65B Dense Open-Weight Model Scoring 80.5 on Terminal-Bench 2.1

    Ricursive Intelligence released Ripple-65B Preview, a 65B dense open-weight model pretrained from scratch on 12T tokens with up to 1M context, shipping three Apache 2.0 checkpoints (instruction-tuned BF16, NVFP4, and BF16 base). Ricursive reports 80.5 on Terminal-Bench 2.1, the top self-reported score among US open-weight models, while trailing larger MoE rivals on SciCode (39.4) and DeepSWE 1.1 (44.3), with full R&D cost stated at $32M. Weights are on Hugging Face, with Together AI, GMI Cloud and OpenRouter listed at launch and an experimental NVFP4 build targeting NVIDIA DGX Spark.

    Why it matters: Useful for readers tracking open-weight agentic coding models, since it pairs self-reported benchmark claims with a detailed architecture, cost and deployment breakdown.

  19. MarkTechPost58

    webAI Releases TwIL-8B, an 8.8B Reasoning Model Scoring 98.2 on MATH-500

    webAI has released TwIL-8B, an 8.79B dense reasoning model post-trained from IBM's Granite 4.2 8B, with weights on Hugging Face under a non-commercial license. It scores 98.2 on MATH-500 and 81.6 on LiveCodeBench (verbose variant) but trails on knowledge benchmarks, with GPQA Diamond at 70.2, 11.5 points behind Qwen3.5-9B. A 4.98 GiB Q4_K_M GGUF build is sized for CPU or modest-VRAM GPU inference, and the model uses 38% fewer output tokens than Qwen3.5-9B on AIME 2026.

  20. Hacker News AI20

    Show HN: Amigent – AI agent and remote desktop for Commodore Amiga over 50 KB/s

    Amigent connects a PC or Mac to a real Amiga over the home network, letting an AI agent such as Claude Code or Codex perform tasks like installing software and changing settings, or letting the user control the Amiga as a remote desktop with live screen, mouse and keyboard, file transfer and shared clipboard. It is tuned for slow hardware and networks, with a 7 MHz 68000 sending its whole screen in under a second and a click over a plipbox showing its result in a fraction of a second. It requires AmigaOS 2.04 or newer, any CPU from the 68000 on, a network connection and a TCP/IP stack, plus a PC or Mac with Python 3.10 or newer, and is MIT licensed.

  21. Hacker News AI65

    OpenRing – self-hosted voice AI agents with a Vapi-compatible API

    OpenRing is an MIT-licensed, self-hosted voice AI agent platform that runs on your own machine or server with your own provider keys and no platform fee, speaking the same JSON as Vapi so clients can migrate by changing the base URL. It runs each call as a Pipecat pipeline (speech-to-text, model, voice) and supports Vapi-style assistants, tools, webhooks, phone numbers, campaigns, knowledge bases, squads and SIP trunks, with a free local stack of Whisper, Ollama and Kokoro. It starts with `npx openring` on Node.js 18 or later, and calls, transcripts and keys stay on your server with no telemetry.

  22. Hacker News AI69

    Show HN: Atlas – Let AI agents use your Mac without taking over your screen

    Atlas is a free, open-source MIT-licensed tool that gives AI agents a hidden second macOS login on your Mac, where they can open apps, click and type while you keep working in your own account. It supports Claude, Codex, Cursor, opencode and Pi, queues agents one at a time, and reports lower context use than native Computer Use (e.g. 36k vs 232k tokens for Claude Opus 5.5 across three tasks). It is not a security sandbox, and some features like native Save/Open panels and Safari web views require a connected Screen Sharing session.

  23. Hugging Face Blog60

    Dharma Field Extractor: Any Document In, Structured Fields in JSON Out

    Dharma-AI introduces the Dharma Field Extractor, a 2-billion-parameter model built on DharmaOCR-LITE v1.1 with a LoRA adapter trained on 43,384 examples, which takes a document image plus either an open request or an exact schema and returns typed JSON, including null for fields not present. On a 546-document, 1,638-case benchmark it scored 0.834 field-level F1, above every frontier configuration tested (Claude Opus 5 at 0.816 to GPT-5.6 Luna at 0.741), while running self-hosted on a single L40S at a cost the company reports as 6 to 147 times lower than the APIs compared. The company frames the result as evidence that a small model trained close to a bounded task can beat a general frontier model on that task.

    Why it matters: Useful for teams weighing self-hosted specialized extraction against frontier APIs, though the benchmark and cost figures come from the vendor itself.

  24. Google Research52

    Agile Deliberation: How AI helps humans define the gray areas

    Google Research introduces Agile Deliberation, a framework that treats building subjective AI classifiers as an iterative conversation, splitting the task into concept scoping (breaking a broad idea into sub-concepts and exclusions) and concept iteration (using vision-language models to surface borderline images that challenge the user's definitions). In 18 live 1.5-hour user sessions across the subjective domains of "Healthy Food" and "Pay-to-Play" content moderation, the approach produced more consistent, higher-performing classifiers than zero-shot, automated collaborator, and manual deliberation baselines while reducing user fatigue and anxiety. The code and algorithms are open-sourced on GitHub, with a Colab/Jupyter demo available.

  25. The Decoder58

    New open-source benchmark reveals reasoning as the biggest bottleneck for AI in cloud operations

    CloudOpsBench is a new open-source benchmark with 45 AWS tasks, built by experienced DevOps engineers, that measures how well AI agents handle typical cloud DevOps work such as fixing services, removing resources, restricting access rights, rolling out updates, and decommissioning projects. Even the best models fail many tasks: Claude Fable 5.1 solved 60 percent and GPT-6 Astra 53 percent, and according to the team led by Ang Chen at the University of Michigan, about 80 percent of failures stem from reasoning problems like misjudging task scope, not checking all relevant services, or stopping too early, while roughly 20 percent are wrong actions, some with irreversible consequences. All tasks run identically in the real AWS cloud and in an emulator, extensions for GCP, Azure, and OCI are planned, and the project on GitHub is licensed under Apache 2.0 with a detailed failure analysis also available.

  26. AWS Machine Learning Blog58

    Four layers of context engineering for production AI agents: How Deep Insight applies them in working code

    This post presents a four-layer context engineering approach for production AI agents—architectural isolation via multi-agent design, prompt-based output control, filesystem-based context externalization, and independent verification with safeguards—illustrated through Deep Insight, an open source multi-agent data analysis solution built on the Strands Agents SDK and Amazon Bedrock AgentCore. Reported results include roughly 95 percent fewer tokens added to context per script from a combined write-and-execute tool and roughly 97 percent smaller system prompts from on-demand skill loading. It is Part 2 of a three-part series.

  27. Hacker News AI18

    Speeding Up eBPF File Opens by 54% (Not AI Gen)

    The author describes three small changes to an open-source eBPF agent (between commits 9c521bd and cfc7538): skipping the LPM cache lookup when a path tail is shorter than 128 bytes, checking the inode cache before building the file path in lsm_file_open, and loading only lsm_inode_permission, tp_exit_openat and tp_exit_openat2 when ptrace blocking is enabled. In a benchmark opening one file 200,000 times and 20,000 new files, repeated opens dropped from 4.68 µs to 2.15 µs (about 54%) and cold opens from 6.33 µs to 5.36 µs (about 15%), with repeated-open p99 falling from 8.3 µs to 3.46 µs. The author notes he runs the perf test after each commit and gives the results to an LLM to find inefficiencies to fix.

  28. Hacker News AI60

    Show HN: Renor – one memory for all your AI agents

    Renor is a shared memory layer for coding agents over MCP that stores decisions, corrections and facts as plain Markdown you own, so what one agent learns in a session is recalled by others in the next. It offers a hosted path via `npx renor connect --client <name>` (claude-code, codex, cursor, grok, gemini, vscode, opencode or pi) and a self-hosted path via `npx -y renor-engine mcp` requiring Node.js 22.16 or later, with notes in ~/.renor and a SQLite search index rebuilt from the Markdown. The engine is Apache-2.0 (1.0.0 was AGPL-3.0), the CLI, SDKs and integrations are MIT, and anonymous telemetry is on by default unless disabled with `renor-engine telemetry off` or DO_NOT_TRACK=1.

  29. Hacker News AI32

    Show HN: Continuum Verifier – offline checks for AI data-provenance records

    Continuum Verifier is an open, offline tool that checks self-contained "capsule" files carrying a content fingerprint, a chain of signed attestation records, Merkle inclusion proofs, the issuer's key, public-witness anchors, and a re-stamp trail, using only the capsule and the repo with no network or issuer access. It returns verdicts of VERIFIED, SELF_CONSISTENT, RESTAMP_NEEDED, WITHHELD, or FAILED, and proves custody and integrity of a record rather than the truth of the underlying facts. It is a v0.1 reference implementation whose sample witnesses and re-stamp signatures are simulated with local keys, with wiring a real durable witness named as the next step.

  30. The Decoder85

    GPT-6 Astra sees the matrix and reprograms other AI models directly through their weights

    Researchers describe "Weight-Space Programming," in which a language model called Astra reads and edits the raw weights of other networks without step-by-step training: it identified which of four games a network was trained to play, generated untrained game-playing weights that beat a standard baseline in Breakout (517 vs. 25 points), and changed 0.016 percent of Qwen3.5-0.8B's weights to raise MATH-500 accuracy from 7.9 to 48.1 percent. It could not measurably improve Gemma 3 270M, and Jürgen Schmidhuber called the approach "old hat," while researcher Omar Khattab said the novelty lies in how well today's language models translate between human language and a network's numerical values. Experiment logs are on GitHub and the edited Qwen model is on Hugging Face.

    Why it matters: It offers a concrete look at weight-level model editing, including benchmark numbers and a public dispute over how novel the idea really is.

  31. Hacker News AI58

    Deterministic security and privacy checks for code from AI assistants

    Secure & Private by Design is an open-source tool that runs deterministic, rule-based security and privacy checks on code written by AI assistants, enforcing the same findings across Claude, GPT, Gemini and human CLI users via editor hooks, git hooks, CI and an MCP server. It detects secrets, Indian PII (Aadhaar, PAN, mobile, email), PHI schema fields, unclassified personal data and missing audit or AI decision logging, and deploys editable scaffolds for consent, data rights, retention, breach notification, RBAC and audit logging. Everything runs locally with no telemetry, and the project states it is not a compliance certification or legal advice.

Nov 4Wed
  1. The Decoder55

    Anthropic's new prototype proves which model wrote an AI response without exposing its weights

    Anthropic released FlashProve, an open-source research prototype that uses zkSNARK cryptography to prove a response came from a claimed model without revealing its weights. For Meta's 400-billion-parameter Llama 4 Maverick, it generates proofs at roughly 110 tokens per second on a single H100 GPU, with each ~19 MB proof verifiable in 17 seconds on a standard CPU, adding 70x to 95x overhead over normal inference. Anthropic notes the prototype only proves contexts up to 1,024 tokens, hides weights but not architecture, and works only when the model picks the single most likely next token rather than randomized sampling.

  2. Hacker News AI38

    Show HN: AI Gude - Personal Podcast News Aggregator

    AI Gude is free, self-hosted open source software that turns the RSS feeds you follow into a personal morning podcast: at scheduled times it reads new articles, writes a script in your style, speaks it with an AI voice and delivers one audio episode with chapters. It uses Gemini for text and speech, Firecrawl for article text, ffmpeg for encoding and SQLite for storage, and supports multiple users via a reverse proxy that sets the X-Aigude-User header. A live demo is available at aigude.alexdoes.dev with the user and password "demo".

  3. Hacker News AI62

    MCP vs. API vs. CLI: How should AI agents connect to your app?

    This post argues that an MCP server should be the first choice for making a web app agent-friendly, because MCP standardizes both server and client behavior, lets the client handle OAuth so the model never sees the token, and can route destructive-action confirmations directly to the human via elicitation. REST APIs remain better for scripts, bulk jobs, data exports and integrations that run without an agent, while CLIs suit agents with a shell working with local files. The author recommends putting business logic in operations and exposing them as thin adapters over MCP, REST and optionally a CLI, using the open-source Linkly link shortener as the example.

  4. Hacker News AI60

    Show HN: Scan-to-vector – rebuild scanned pages as vector PDFs with an AI agent

    Scan-to-vector is an agent skill for Claude Code, Codex and other clients that read SKILL.md, which rebuilds scanned pages as searchable PDFs with text, vector-redrawn diagrams and charts, and labelled raster crops, checking each page against the scan via a ledger and overlay review. The author reports benchmarks on unseen documents where all four tested models scored higher with the skill on long documents (Haiku 5.5 from 0.841 to 0.997, Sonnet 5.5 finishing 37% faster), while noting the benchmark is synthetic printed English pages and the workflow takes minutes per page. It is installed via a Claude Code plugin or install.sh, requires Python 3.10+ with PyMuPDF, Pillow, numpy, scipy and tesseract, and is released under the MIT License.

  5. Hacker News AI58

    Show HN: Shield KYA – Open-source audit trail and gate for AI agent tool calls

    Shield KYA is an MIT-licensed CLI and local MCP gate that records every AI coding agent tool call on a live local dashboard and checks calls against a policy baseline, returning Allow, Review, or Deny. The open-source client is advisory and fails open except for a small never-list of dangerous calls, while real blocking and human approval run on the hosted MCP policy path. It installs via npm (@shield-agent/kya) and requires Node.js 22+, with plugins for Claude Code, Copilot CLI, Codex, and Gemini CLI.

  6. Hacker News AI75

    Show HN: Umbra – An AI agent that uses Android apps while you use your phone

    Umbra is an Android agent that reads the screen via the accessibility tree and screenshots, then taps, types, scrolls and opens apps in an invisible background display so you can keep using the phone. It also runs a Debian Linux environment on-device for building and testing code, supports Anthropic, OpenAI, Google, OpenRouter or any OpenAI-compatible model, and requires your approval for risky actions like payments, messages or deleting data. It requires Android 16 (API 36), root is optional, and distribution is by sideload from umbra-agent.pages.dev.

  7. MarkTechPost69

    Tencent WeVisOne-VL Unifies Detection, OCR, Segmentation and Matting

    Tencent's WeChat Vision team released WeVisOne-VL, an open vision language model in 2B and 4B sizes that handles detection, referring, counting, pointing, OCR, segmentation and alpha matting in one model, with pixel outputs from a 9.6M-parameter decoder so no separate SAM is needed. Built on Qwen3-VL, it is trained with up to 8,192 tokens of context and runs on a single GPU via Transformers or vLLM in BF16. Tencent reports the 4B model scoring 71.3 F1 on Dense200 and 81.8 F1 on RefCOCOg test, while noting a 52.6% rejection rate on PR-Bench.

  8. Hacker News AI40

    I built a free local AI co-processor in Rust. AXON Developer Edition

    AXON Developer Edition is a free, local AI co-processor built in Rust that runs on Windows 10/11 and lets AI assistants access files, memory and tasks on the user's own machine without sending data to the cloud. It is not an AI model itself, and users connect assistants such as Claude via its MCP connector or other web-based AIs through AXON LOCKA, a temporary authenticated tunnel. The full build requires no account and is downloaded as a ready-to-run binary that listens on port 8880.