#开源/仓库

Show HN: AI agent needs process to build a commercial application

NoMoreHappyPath is a free, open-source (Apache 2.0) delivery system for macOS and Linux that runs AI agents through 21 quality steps, with a builder, an independent reviewer from a different AI company, and a CTO, in a plain-English control room. Tests are written before code, must fail first, and are run by a board agents cannot edit, with nothing accepted until the user approves it. It runs on the user's machine with their own AI accounts, and the author cites a weather app built in 48 minutes with four reviews, one of which failed.

Show HN: Navital Gym – open environments for testing AI agents on Indian tasks

Navital Labs released Navital Gym, a public, actively developed suite of six sandboxed, deterministically auto-graded environments (trip planning, GST invoicing, forms, multi-document forms, Hindi/Hinglish customer support, and UPI) for testing AI agents on Indian-reality tasks without an LLM judge. It runs on plain Python 3.10+ with optional adapters for Inspect AI, verifiers and OpenEnv, and publishes pass@1 and partial-credit results for small, mid-size and larger models. The environments use simplified, documented rules and synthetic data and are not affiliated with NPCI, RBI, GSTN/CBIC, IRCTC or any government body.

Show HN: Dodaiso – Open-source personal AI agent (like Jarvis, dots,duet)

Dodaiso (formerly Personal Jarvis) is an open-source, Apache 2.0 agent OS for desktop that runs coding agents, browser and apps by voice or chat on your own machine, with no account or paid tier and support for hosted, subscription or local models. It offers an Agentic IDE with multiple coding agents in live panes addressed by call signs, plus agents, memory, plugins/MCP, a marketplace and approval checks on system-changing actions. It runs on Windows, macOS, Linux or a headless server, installable via a one-line script or desktop installers.

Show HN: AI agents trade privately on BitcoinZ, with proof of delivery

This Show HN post introduces I Am Wallet, an MCP/REST service that lets AI agents create shielded BitcoinZ wallets, send signed private messages, and trade in small paid steps where each delivery is fingerprinted and checked. It describes built-in safety limits, Ed25519 signer identities, a 1 BTCZ service fee per payment or message, and honest caveats including custodial key storage and reliance on current zk-SNARK cryptography. The code is MIT-licensed and can be self-hosted with a synced BitcoinZ node, Python, and nginx.

Edge0 Releases Audio8-TTS-Realistic, a 0.6B Voice Cloning TTS Model

Edge0 has released Audio8-TTS-Realistic, a 0.6B-parameter open-weight text-to-speech model for zero-shot voice cloning from one reference clip and transcript, supporting 23 languages under Apache 2.0. It uses a dual-autoregressive design on Qwen2.5 0.5B and a three-stage training pipeline whose final GRPO stage is guided by a learned realism critic. Edge0 reports 4.86% average error on CV3 across 9 languages, ahead of Qwen3-TTS-1.7B (7.73%) and VoxCPM2 (5.35%), though larger models like Higgs TTS 3 still lead and Thai WER is near 49%.

Show HN: Breatic – Collaborative canvases and documents with an AI agent

Breatic is a real-time collaborative AI workspace where creative teams combine Document Spaces for briefs and scripts with Canvas Spaces for generating and refining media assets, plus an AI agent that works within project context. It is available online and source-available under the Breatic Source-Available License v1.0, which permits individual, private-group and single-organization internal use but prohibits public offerings without separate authorization. Self-hosting uses Docker images with PostgreSQL and Redis, plus a Cloudflare Worker and R2 bucket for uploads, and is not an offline or wholly on-premises installation.

Baseten Releases joey-1, a 2B Decision Model at $0.01 per 1M Tokens

Baseten's Base Labs released joey-1, a 2B decision model post-trained from Qwen3.5-2B that returns a label and calibrated probability for choice, yes/no and score questions instead of generated text, priced at $0.01 per 1M input tokens with MIT-licensed weights on Hugging Face. Baseten reports 20 ms P50 latency for an average 2,000-token request and a score of 63.30 on the Jev Decision Index 0.4 public split, ranking #37 of 230 models and top under 4B, though it trails its 9B sibling elephant-1 by 8.94 points. It runs on Baseten Model APIs, dedicated deployments (262k context) or self-hosted on 1 CUDA GPU, and is also listed on OpenRouter.

We ran AI agent swarms in Minecraft and Mindustry

The authors partnered with Olam Labs to run multi-agent experiments in Minecraft and Mindustry, building and open-sourcing a long-running Minecraft agent harness on Mineflayer with a Pi-driven agentic loop. In a 3-hour run, 50 GPT-6.1 Sol agents pursuing "Build a thriving civilization" formed a hierarchy led by Ada, wrote shared plans and rules, sent about 5,900 messages, and shifted from exploration to food rationing after 22 of the run's 34 deaths occurred on the first night.

Prime Intellect Releases Prime Mega MoE: Trainable Fused MoE Kernel, 3.11x Faster Forward Pass

Prime Intellect released Prime Mega MoE, a training extension of DeepSeek's fused Mega MoE kernel that adds an overlapping backward pass, making the kernel usable for training rather than inference only. It reports 32% to 34% faster full training steps than the prime-rl baseline and 48% to 55% faster than TorchTitan on Qwen3-30B-A3B and Qwen3-235B-A22B, with a best case of 3.11x faster forward pass on a Qwen3-30B-A3B MoE layer. It ships in prime-rl as the mega_moe dispatch backend, with kernels in a prebuilt DeepGEMM wheel from prime-kernels.

AI Memory on Memgraph: Persistent Context Across Session, Team and Organization

Memgraph describes Context Graph, part of its AI Toolkit, which stores agent session activity in Memgraph and runs post-session reconciliation using the local GLiNER model to create memories, entities and relationships linked back to the raw record. Because memory lives outside any single agent harness, a later session in a different tool can recall context such as a prior test_login failure caused by the shared token cache in auth-service, owned by Ana. Memgraph Enterprise adds role- and label-based access control, multi-tenancy and auditing for shared deployments, and a community call on October 29 covers building your own AI memory.

Show HN: Kinoloom – A local AI film studio for an 8 GB GPU

Kinoloom is a one-person Windows project that runs large open video models (LTX-2.5 at 22B parameters and MiniMax H3 at about 20B) on a single 8 GB RTX 3070 by giving the card one model at a time and streaming the rest from 64 GB of system RAM. It wraps the models in a local web app with an LLM story writer, a story bible, Qwen-Image 2.1 cast portraits and opening frames, a seamless-chain engine that joins short clips in latent space before decoding, face-identity QA via InsightFace, and an assembler that cuts the film. The author reports about 6 minutes for a warm 5-second 1080p LTX-2.5 clip and 20-22 minutes for MiniMax H3, notes the project has been tested on only one machine, and does not publish H3 renders because the H3 Community License restricts use and output display in the EU, UK, South Korea and the US.

Mistral releases weights for its trillion-parameter AI model, targeting cybersecurity

Mistral has released the weights for Mistral Large 4 (ML4), nicknamed "Le Chonk," a trillion-parameter open-weight model that businesses and governments can download and deploy themselves, available on Hugging Face and via its API. Mistral says the model is comparable in performance to Chinese open-weight competitor GLM-5.3 while using 47 percent fewer tokens per task, and claims it outranks all other open models on Artificial Analysis' Coding Agent Index for agentic coding tasks. The company positions ML4 for sovereign cyber defense, with VP of Science Pierre Stock saying open weights give organizations access to frontier intelligence "that can't be cut off by definition."

Featured coverage of this story: 《Le Chonk: Built for Coders, Trusted by Defenders.》

Show HN: Sheetgraph, AI agents can understand complex spreadsheets

Sheetgraph is a Rust-built local binary that extracts Excel and Google Sheets files into a short REPORT.md plus a formula dependency graph, keeping every cell for exact lookups; a 120k-cell invoice ledger compresses to about 7k tokens instead of roughly 1.3M. It ships as an agent skill for Claude Code, Codex and Cursor, and supports drill-down commands (cells, trace, find) for querying specific values. It is in beta (0.1), and the report format and library API may change between releases.

Show HN: phpClaw – AI agent runtime for PHP with tools, memory and guards

phpClaw is an MIT-licensed AI agent runtime for PHP that provides a built-in ReAct loop, pluggable tools, memory, and eight default security guards, with zero framework dependencies in core. It auto-detects seven LLM providers from environment variables, ships eight framework and CMS adapters (Laravel, Symfony, Drupal, Magento, WordPress, Joomla, OpenCart, PrestaShop), and includes an MCP server for exposing tools to Claude Desktop and Claude Code. It also offers pipelines and graphs with human pause/resume and fan-out, and requires PHP 8.1–8.4 with ext-curl, ext-json, and ext-mbstring.

Decisions for a penny: introducing joey-1, our fastest decision model

Base Labs introduces joey-1, a decision model with 3B or fewer parameters that returns a label and calibrated probability score in a single forward pass, serving an average 2,000-token request at a P50 end-to-end latency of 20 ms (about 2.5x faster than elephant-1) at $0.01 per million tokens. Post-trained from Qwen3.5-2B using the same data and recipe as elephant-1 and mammoth-1, it is available on Baseten Model APIs with open weights on Hugging Face, and supports a 16k context window on Model APIs versus the full 262k on dedicated deployments.

Luma – smart launcher for you, and your AI agent

Luma is a Linux launcher that fuzzy-searches apps and files, offers opt-in CLIP semantic photo search and OCR text-in-image search, a calculator and currency conversion, and exposes the same photo and screenshot index to AI agents via an MCP server (luma mcp) and CLI (luma query). It runs locally by default, with only model downloads and exchange rates leaving the machine, and supports Hyprland, niri, GNOME and KDE. The page reports one agent prompt taking 11 s with Luma versus 1 min 26 s without.

Paris-based Mistral releases "Le Chonk," a strong coding model that still trails China overall

Mistral released the weights of Mistral Large 4, nicknamed "Le Chonk," a roughly one-trillion-parameter multimodal model (52 billion active per step) that it says leads open models on coding, scoring 41 percent on Terminal-Bench 4.0 and 19.7 percent on FrontierCode 2.0 while trailing Claude Opus 5.5. Independent testing by Artificial Analysis put the final version at 26 points on its Intelligence Index, behind Chinese open models DeepSeek V4.1 Flash and GLM-5.3 at 32 and MiMo-V2.6-Pro at 30, though it is fast at 161 tokens per second and cheaper per task than the preview. The model is under a modified MIT license requiring a separate license for commercial cloud hosting, and API pricing is $0.70 per million input tokens and $2.10 per million output tokens.

Show HN: Howler, a macOS menu bar AI that burns the chat after reading

Howler is a macOS menu bar AI app that opens a parchment-style letter with ⌥H, streams answers at a hand's pace, and burns the conversation when you press Esc or click outside, saving nothing. It supports chat-completions, Anthropic and Gemini formats with configurable base URL, model, body and headers, and stores only the model setup, prompt instructions, shortcut and API key in the Keychain. It is signed with a Developer ID and notarized by Apple, self-updates via Sparkle from v0.1.7, and is MIT licensed.

Show HN: I built a tamper-evident flight recorder for AI agents

Mnestiq is a flight recorder for AI agents that records what an agent read, decided and did, tags each piece of model context by provenance, and hash-chains records sealed by Ed25519-signed Merkle checkpoints so edits, deletions and reordering fail verification. It can be used without code changes via a proxy between the agent and model APIs, or through Python SDK wrappers for Anthropic and OpenAI clients, and includes a local dashboard with detection rules such as MNQ-001 for untrusted arguments in tool calls. The project is at version 1.0 under Apache-2.0, with a documented evidence format and a stated security model noting that evidence is tamper-evident after writing but a compromised host is not made truthful at write time.

Show HN: Apargon – Ask AI for any app and run it safely with a double-click

Apargon is a macOS and Linux desktop app that runs "papps"—single HTML files with extra <meta> tags—in a sandboxed window with no installer, server, or build step. The sandbox blocks anything the tags don't allow, and a popup shows requested permissions before the page runs; a coding agent can write a papp from one prompt. Install is via a curl script, release downloads, or Homebrew, and Windows is not yet supported.

Day 0 support for Mistral Large 4 on vLLM

vLLM announces day-0 support for Mistral Large 4, a 1T-parameter MoE multimodal model with 49B active parameters per token, running on the V2 model runner with breakable CUDA graphs and a new vllm/models/mistral/ package. The work adds cross-layer KV sharing, DCP/PCP context parallelism, DSpark speculative decoding, NIXL-based disaggregated serving and HiSparse KV offload, plus day-0 ROCm support on AMD Instinct MI350 Series GPUs with MXFP4 experts. Support ships in the vllm/vllm-openai:mistral-large-4 Docker image, with FP8 and NVFP4 checkpoints and serving commands documented.

Show HN: SpaceApp – Self-hosted workspace running AI coding CLIs in parallel

SpaceApp is an MIT-licensed, self-hosted workspace that runs up to 16 AI coding CLIs (including OpenCode, OpenAI Codex, Claude Code, Gemini, Qwen, Kimi, Grok and DeepSeek) side by side in persistent rooms on your own machine, installed with a single npx command on Linux, macOS or Windows 11. It adds a coordinating agent called Nous, optional copilots and studios, a Vault for logins, and a Space Gateway that uses your subscriptions before API keys. The post notes that rooms are not a security boundary between mutually untrusted users and links to a Known Limitations page.

Le Chonk: Built for Coders, Trusted by Defenders.

Mistral released the open weights of Mistral Large 4 (Le Chonk), a 1-trillion-parameter natively multimodal model trained from scratch in Europe, reporting leading open-weight results on coding benchmarks such as FrontierCode 2.0 (19.7%) and Terminal-Bench 4.0 (41%), plus 51% on the Artificial Analysis Cyber Index. It offers three reasoning modes (Low, Medium, High) and is priced at $0.70 input / $0.07 cached input / $2.10 output per million tokens under a modified MIT license that requires a separate commercial license for Model-as-a-Service offerings.

Sorry Harvey! I Made a Free Open-Source Alternative to Your Overpriced Legal AI

The post introduces CaseCite, a free, open-source legal AI platform released under the Apache 2.0 license that handles cited legal research, document analysis, contract review and redlining, legal drafting, citation checks, and judge intelligence. It is self-hosted, runs on the user's own infrastructure, and supports the user's own AI models and API keys, with users paying only their own infrastructure and AI usage costs. The author acknowledges it is not yet as polished as Harvey's enterprise platform but says it covers a substantial portion of everyday legal workflows.

Show HN: Crewd NPM-style package manager for AI coding agents and skills

A Show HN post introduces Crewd, described as an NPM-style package manager for AI coding agents and skills. The submission links to its Hacker News comments page and had 1 point and 0 comments at the time.

Show HN: Vibecheck – a pre-launch security audit for AI-built apps

Vibecheck is a zero-dependency scanner with 60+ checks plus AI-agent skills that audit AI-built apps for exposed secrets, missing Supabase RLS, auth gaps, and deploy misconfigurations, producing a 0–100 Launch Score with a GO/NO-GO verdict and copy-paste fix prompts. It includes an always-on guard hook that blocks agents from hardcoding API keys or committing .env files, and can run in CI via a single Node script with --fail-on critical. The audit is read-only, makes no network calls, and never runs the project's code.

Project Baobab – Off-grid solar AI edge nodes with micro-payout incentives

Project Baobab is a lightweight, decentralized edge-computing orchestrator that shards AI tasks across distributed low-power edge nodes such as Raspberry Pi 5 and off-grid hardware, settling compute rewards via automated micro-payments. It uses Python's concurrent.futures for task sharding and the PayPal REST API v2 (OAuth 2.0) to trigger payment orders upon verified task completion, targeting solar/battery-powered nodes. The project is built in Python 3.10+ and runs cross-platform on Linux, Debian, Ubuntu, and Raspberry Pi OS.

Show HN: The AI workspace for teammates share agents on company

Nakama is an AI agent platform for teams, offering a ChatGPT-like dashboard where each agent has its own tools, skills, memory, identity, and permissions, plus channels including Telegram, WhatsApp, Discord, and Slack. It supports multi-org servers with invites and roles, and can be deployed via managed cloud or a single Docker container. Integrations include Composio for over 1,000 external apps, and the project is MIT licensed.

Open-sourcing Deriv One: the self-hosted AI platform we run in production

Deriv has open-sourced Deriv One, an MIT-licensed, self-hosted AI platform combining a reviewed skills catalogue, registered Slack agents, a multi-pane workspace, sandboxed code execution, and a no-code app builder, all running on one identity directory and secrets vault with usage tracked against the company's own model gateway. The team says the real problem was never discovery but control—deciding who can run which skill—and that 221 skills came from 54 people across the company. The repository includes provider interfaces, Docker images, Kubernetes starter manifests, and adopter documentation, but excludes Deriv's production cloud plugins, internal integrations, Terraform and Helm configuration, and Slack DMZ relay implementation.

Introducing AnySpeechRecognizer and AnyDecisionModel: Open Models for Speech and Decisions on Apple Platforms

The author announces two Swift packages: AnySpeechRecognizer, which mirrors Apple's SpeechAnalyzer API while allowing alternative engines such as NVIDIA's Parakeet and Alibaba's Qwen3-ASR running locally on Core ML, plus speaker diarization via NVIDIA's Streaming Sortformer; and AnyDecisionModel, which answers yes-or-no, multiple-choice, and rating-scale questions with typed results and probabilities from backends including MLX, Core AI, llama.cpp, Jev, and Ollama. Both can download open models from the Hugging Face Hub and run them on device, require Swift 6.2 and macOS 26 or iOS 26 or later, and are published on GitHub.

AWF – AI Workflow Framework

AWF is an open-source ecosystem of three binaries: the Go-based AWF CLI for provider-agnostic YAML workflows, ZPM (Zig + Prolog) for deterministic logical reasoning via 22 MCP tools, and ztick, a push-driven Zig scheduler that can dispatch into AWF workflows. The project emphasizes deterministic replays, per-step token tracking, provider swapping via YAML, and static binaries with zero runtime dependencies.

Infini-AI-Lab Releases Hermit, an Open-Source Harness Adapter Framework for AI Agents

Infini-AI-Lab has released Hermit v0.1.0, an Apache-2.0 Python framework that trains an external "harness adapter" while the agent's model and harness stay frozen, letting Codex or Claude Code agents adapt to new tasks without fine-tuning or forking the harness. It ships five evolution methods (GEPA, SkillOpt, MetaWorld Model, Harness², Continual Harness) and reports gains over no-adapter baselines on OfficeQA, CL-Bench and ServeLearnBench with GLM 5.3 Flash, including 0.795 accuracy on CL-Bench database exploration versus a 0.580 in-context baseline. The results are self-reported and mostly single runs, and adapters can be deployed to Codex, Claude Code or Pi or exported as native plugins.

Open Agent Primitives: Run enterprise AI agents securely

Open Agent Primitives (OAP) is a platform for running enterprise AI agents in your own cluster on models you choose, where the platform rather than the model checks every action before it runs. It ships implementations of six primitives—agent definitions, safe tools, credential brokering, authorization, channels and memory—with controls such as per-action permission checks, injected credentials the model never sees, and signed hash-chained audit logs verifiable offline with `oap audit verify`. The material also maps OAP's coverage against the OWASP Top 10 for Agentic Applications, listing four substantial and six in-progress items with remaining gaps noted.

Show HN: Mediary Scout – Open-source AI agent for cloud drive media tracking

Mediary Scout is an open-source AI agent that acquires media into personal cloud drives (Quark, UC, 115, 123Pan, GuangYaPan, Tianyi) instead of local NAS storage, using a write-scoped sandbox and post-transfer file verification rather than regex rules. It coordinates multiple drives with single-drive priority and cross-drive gap filling, blocks over-quota drives, deduplicates staged files, and can fetch Chinese subtitles from assrt.net. It ships as a signed macOS/Windows desktop app or a Docker Compose stack with a Next.js UI, PostgreSQL, and a bundled PanSou search instance.

Shellfirm – a seatbelt for AI agents running shell commands

Shellfirm is a shell hook that intercepts dangerous commands before they run, offering 496 built-in checks across 23 groups, context-aware escalation (SSH, root, protected branches, production clusters), safer alternatives, and an audit trail. It supports hooks for 8 shells, team policies via .shellfirm.yaml, and an MCP server plus hooks for AI agents such as Claude Code and Cursor. The project is Apache-2.0 licensed and installable via npm, Homebrew or Cargo.

Bernstein – open-source governance layer for AI agents

Bernstein is an open-source (Apache 2.0) governance layer for AI agents in which a deterministic Python scheduler runs agents in parallel with no model in the coordination loop, so the same plan replays byte-identically. It enforces policy as code across four stages—decompose, spawn in isolated worktrees, verify via tests/types/lint and optional cross-model review, and merge only verified diffs—and writes an HMAC-chained audit log plus an optional Ed25519-signed run receipt that can be verified offline. The material cites v3.23.0 with 53 CLI adapters and describes Cloudflare Workers, Kubernetes and on-prem deployment options.

WSP MCP – Free WordPress AI Agent Plugin

WSP MCP is a free WordPress plugin that connects MCP-capable AI clients such as Claude, ChatGPT, Cursor, Codex, OpenClaw and OpenCode to a WordPress site, offering 200+ tools with no paid tier or companion plugin. Write abilities are off by default, each tool enforces WordPress capability and per-object permission checks, and every call is recorded in a built-in Audit Log stored in the site's own database. Version 2.9.6 hardens authorization checks on post listing, Elementor, Ultimate Addons for Elementor and media-upload tools, and requires WordPress 6.9+ and PHP 7.4+.

A zero-dependency auditor for AI agent exposure (secrets, SSH, configs)

Dante is a free, MIT-licensed Node.js command-line tool that checks a machine for exposed secrets, risky SSH and Windows settings, and AI agent configuration, showing the proof line behind each verdict and reporting UNKNOWN rather than SAFE when it cannot read the evidence. It has zero dependencies and no install scripts, runs offline by default, and only uses the network for three actions (knowledge pack fetch, catalog, optional bait watch) after the user approves a printed command. The page argues signature antivirus is the wrong tool for AI agents, which run with the user's rights, citing the August 2025 s1ngularity attack on Nx and the later Shai-Hulud npm worm as examples of credential-hunting that worked because secrets were readable.

Show HN: Fengarde–OCSF-native open-source SIEM with local AI triage (Apache-2.0)

FENGARDE is an Apache-2.0, OCSF-native open-source SIEM that turns a raw log line into a verified incident and a draft NIS2 notification, built as nine services communicating only over a bus. The repository includes a demo where a real SSH brute force and DNS burst sent over UDP 5514 become a real alarm, plus a console with twelve views and a capability board marking anything stubbed or not built. The UDP syslog listener is unauthenticated, so port 5514 should be exposed only on a trusted network or VPN.

AI agent skills for better second-hand decisions

A GitHub-hosted set of portable agent skills (Buy, Kits, Inspect, Collect, Keep, Second Look) helps assistants research second-hand purchases by separating seller claims, product documentation, and unverified facts. The skills produce source-backed shortlists, compatible bills of materials, inspection cards, collection plans, and repair-versus-replace comparisons, and they do not contact sellers, bid, or purchase. Install via `npx skills add budna-marketplace/recommerce`; Budna MCP is an optional integration.