OPEN SOURCE

AutoGPT's founding AI engineer published the gates that keep an open source repo sane when agents submit the majority of pull requests: enforced PR templates, AGENTS.md placement, skills that fire on trigger phrases, a CLA as a human detector, and a commit-SHA rule that kills fake review resolutions. GitHub published the playbook August 12, and the details are sharper than the headline.

EvoX Genesis built a 250k-line Rust C compiler with DeepSeek V4 Flash for $44 in tokens by making the project the persistent thing and keeping agents finite-lived. The paper's three runs, the design that made them possible, and what it says about agent memory.

Liquid AI released LFM2.5-VL-3B on August 12, 2026: a 3.1B open-weights vision-language model that averages 80.7 on ScreenSpot-v2, doubles ToolSandbox to 59.5, and decodes at 228 tokens/s on an M5 Max in about 3 GB of memory. Here is what shipped, the benchmark caveats, and how to run it.

Cactus open-sourced Needle 2, a 45M-parameter agentic LLM in a single 14MB binary that runs a full tool-calling session in 28MB of RAM. 500 tok/s on a Raspberry Pi 5, ESP32-S3 class parts, Apache 2.0. Here is what the benchmarks actually show.

Multiverse Computing open-sources two changes that make knowledge distillation cheap enough to run at scale: caching the teacher's top-100 logits once so it never sits in memory during training, and a fused chunked KL loss that never materializes the vocab-by-sequence matrix. A GPT-OSS 20B distillation at 32K context drops from four GPU nodes to one, with step time down 5x.

Meta open-sourced Muse Glimmer, a 30B Apache 2.0 multimodal agent model that runs in a 24GB envelope at up to 233 tok/s. MCP Atlas 75.5, SWE-Bench Verified 76.0, 131K context. Here is what the numbers actually say.

A developer's Claude-built night-sky site reproduced an open source project's name, feature set, and even a bug the author had already fixed. The saga that followed says a lot about memorization, accountability, and the verification duties of AI-assisted shipping.

WeatherNext Cyclones adds a full day of lead time to tropical cyclone forecasts - roughly a decade of meteorological progress - and now the weights, code, and data feeds are public. What the paper actually shows and how to run it.

OpenJDK's interim policy bans AI-generated contributions in full or in part, while Oracle runs on AI-written code internally. What the policy actually says, how it compares to Rust and Debian, and what it means for Java contributors.

On August 5 Cloudflare open sourced Cloudflare OS, the agent workspace it has run internally since May: capability-based Gatekeepers instead of ambient MCP access, apps as private per-user instances, and approvals that simulate outcomes so agents never stall. A concrete blueprint for the company-wide agent platform.

Liquid AI shipped LFM2.5-2.6B on August 4, 2026: a 2.6B open-weight model trained for agentic work inside real harnesses, decoding at 220 tokens/s on an M5 Max and 113 tokens/s on a Ryzen CPU. Here is how it was trained, what the benchmarks say, and how to run it.

Prime Intellect open-sourced Prime Agent on August 5, 2026. It gives the model exactly one tool - a persistent IPython kernel - and lets the harness rewrite its own prompts, skills, memory, and sub-agents mid-run. Here is how it works, what the benchmarks actually show, a full provider and model guide, and an honest comparison to Claude Code, Codex, OpenCode, OpenClaw, Hermes, and Pi.

Shieldstral is a 3B-parameter Apache 2.0 multimodal safety classifier that takes your moderation policy as a plain-language question at inference time, scores content 0-1 in a single forward pass, and runs on one 16GB GPU. It beats 12B-20B guard models on text safety and sets state of the art on multimodal benchmarks.

Steve Yegge's response to AI-generated pull requests suggests a better maintainer workflow: automate triage, repair good ideas, and keep human taste at the boundary.

A ripgrep musl binary crashing during very-large searches turned out to be a suspected Linux 7.0 kernel race - a thread's own store vanishing mid-function. The reporter's instrumentation pinned it, and a kernel-hardening maintainer posted a one-line fix candidate for testing.

An analysis from the Earendil team behind the Pi harness documents how OpenAI, Anthropic, and Google now return provider-sealed state instead of portable transcripts - encrypted reasoning blobs, opaque compaction, hidden subagent messages. The five tests and seven rules for session portability, and why session lock-in matters more than model lock-in.

A new 106-issue benchmark across 49 repositories finds frontier coding agents rarely retrieve AI contribution rules on their own - and never refuse to contribute in AI-banned repositories, no matter the prompt. Disclosure and verification can be fixed; bans cannot.

DeepSeek shipped the official V4 Flash release on July 31, 2026. The re-post-trained 0731 build beats V4-Pro-Preview on agent benchmarks at $0.14/$0.28 per million tokens. Here is what changed and how to run it through OpenCode today.

GitHub open-sourced casefold, a Rust crate that folds the case of every byte Blackbird indexes at memory bandwidth. The counterintuitive trick: delete the early-exit, kill the branches, and fold Unicode as byte arithmetic.

Inkling-Small is a 276B-parameter MoE with 12B active per token, Apache 2.0, and open weights. It beats the 975B Inkling on SWEBench Verified (80.2), HLE (31.6), and tool use at a quarter of the size and a third of the output price.