Developer Tools

Meta

2026-10-03 01:29:10

Meta open-sources Muse Gadgets and rolls out Home Link for custom AI hardware

Meta has expanded its Muse push beyond software with the launch of Muse Gadgets, an open-source hardware initiative that lets developers build custom peripherals for its personal AI agent. According to the company’s description, developers can run Muse firmware on ESP32 boards or use a Linux SDK to connect more capable devices, including Raspberry Pi 5-class edge hardware. Meta also outlined several reference designs, such as color E Ink displays for schedules and reminders, HDMI display sticks for TVs and monitors, and handheld touch devices that work as portable Muse terminals. At the same time, Meta introduced Muse Home Link, a small USB-C-powered device designed to connect Muse to a home network and communicate with smart home products, including TVs, speakers, and any device that supports HTTPS interfaces. The company said the first production run totals 5,000 units, with shipments expected to begin in the coming weeks. Eligible Muse subscribers will be able to receive the device for free while inventory lasts. Meta framed the move as a way to let developers experiment with new hardware forms around AI agents. Early examples have already appeared, including an E Ink companion display built by Federico Viticci and a wine-cellar tool created with a roughly $50 SenseCAP Watcher device.

40

Arkham

2026-10-02 11:32:43

Arkham updates API docs with role-based usage guides

Blockchain intelligence platform Arkham has updated its API documentation and added step-by-step guides organized by user role and use case, according to Techub News. The new section arranges multiple API calls by task flow and shows requests, responses, and key information points. Arkham said the update is designed to help different types of users complete practical work more efficiently. The update was referenced in a post from @arkham.

70

BNB Chain

2026-10-01 00:06:48

BNB Chain adds six audit firms to AvengerDAO security marketplace, bringing total to 17

BNB Chain said six security auditing firms have joined its AvengerDAO security marketplace: QuillAudits, Cyfrin, Asymmetric Research, MixBytes, Consensys Audits, and Pashov Audit Group. With the additions, the number of partner security companies on the marketplace has risen to 17. The marketplace is designed to help developers review available services, reach out to audit providers directly, and seek onboarding guidance from the AvengerDAO management team. According to the announcement cited by Techub News, the move is intended to expand security support for builders in the BNB Chain ecosystem while lowering the barrier to development.

60

Pi

2026-09-30 05:16:35

Pi adds built-in MCP in version 0.99.0 and routes it through Codemode

AI coding tool Pi has formally added built-in support for MCP in version 0.99.0, while also changing the wording on its website from "No MCP" to "Now with MCP+Codemode." The move stands out because Pi had previously criticized the standard way MCP is commonly connected to models. Mario Zechner, the author cited in the report, said tests he ran last year showed that the descriptions for 21 tools in Playwright MCP used about 13,700 tokens, while 26 tools in Chrome DevTools MCP used about 18,000 tokens. In his view, exposing too many tools wastes context, makes composition harder, and pushes large volumes of returned content directly into the model context. Pi's new setup keeps the general MCP interface but changes how tools are surfaced. Instead of exposing all MCP tools to the model by default, Pi hands them to Codemode. The model can then write JavaScript to call tools in parallel, combine and filter them, and send back only the results that are needed. Tools can also be loaded later instead of being declared up front. Pi team member Earendil has not withdrawn the earlier criticism, saying many MCP servers still pack in large numbers of tools and text outputs, and that the tool-combination issue has not been fully solved.

210

OpenAI

2026-09-30 03:01:48

OpenAI updates Codex Cloud, letting local jobs continue in the cloud

OpenAI rolled out a broad Codex update at DevDay, with Codex Cloud now able to save a reusable cloud development environment for each project. That setup can include the code repository, dependencies, tools, and permission settings, allowing developers to reuse the same environment across tasks instead of rebuilding it each time. The update also changes how work moves between local and cloud environments. Developers can start tasks directly in the cloud, or hand off work already being processed by local Codex to Codex Cloud. Once launched, the task continues in the cloud workspace, where Codex can keep editing code and running tests without relying on the local computer. Results can later be checked from desktop, web, or mobile, and changes can be synced back to the local machine. OpenAI also added voice control to Codex CLI and introduced a new /agents page for assigning work to multiple agents at the same time. On desktop, Codex now includes Code Review support for GitHub pull requests and GitLab merge requests, with first-pass reviews able to run in the cloud. Codex Security Cloud was also expanded with automated security checks for new commits and scheduled scans across full GitHub repositories. Daybreak Blue is now included by default and no longer requires a separate application.

260

Cloudflare

2026-09-29 06:34:21

Cloudflare expands cf CLI for AI agents as usage reaches 48% of Wrangler

Cloudflare has significantly expanded its new <code>cf</code> command-line interface, a tool first introduced in April, and is positioning it to eventually replace Wrangler. The company said the CLI has grown from supporting only a small number of products to covering more than 3,000 Cloudflare API operations. By comparison, Wrangler currently offers about 280 command paths. The new tool is built with coding agents in mind: command results return in JSON by default, making them easier for systems such as Claude Code and Codex to read directly. Cloudflare also added <code>cf cli search</code>, which lets an agent describe a task in natural language and then locate the matching command without prior knowledge of the company’s API surface. Cloudflare disclosed that agent usage has climbed sharply, rising from about 25% of Wrangler usage in March to 48% over the past week. It also said agents use nearly twice as many distinct commands as humans on average and are nearly four times as likely to execute more than six commands. The configuration model has also changed, with <code>cf</code> adopting a TypeScript-based <code>cloudflare.config.ts</code> file and a <code>cf migrate</code> path for existing projects.

540

Jevgrep

2026-09-28 10:43:08

Jevgrep targets code search for coding agents, with SWE-bench run showing a 28.63% cost drop

Developer David has open-sourced Jevgrep, a research tool built on Jev to help coding agents find relevant code before handing the context to tools such as Claude Code and Codex. The setup is aimed at the code-search step, which the project says is one of the most token-intensive parts of agent workflows. Rather than feeding an entire repository into a model or relying on a single semantic search pass, Jevgrep searches a codebase in stages, decides which directories are worth exploring, inspects related files and declarations, and returns source snippets, line numbers, and clues for follow-up reading. The repository also includes a Skill that tells an agent when to call the `jg` command to gather context. In the latest SWE-bench test covering 10 Python tasks, runs with and without Jevgrep both solved 8 tasks, but total GPT-5.6 Sol cost fell from $7.62 to $5.44, a 28.63% reduction. The author had initially posted 40% on X, but the repository later updated the result to about 30%. That figure only covers Sol costs. Logged Jev calls account for at least another $1.57, while some calls do not have complete billing records, leaving the actual total unknown. The test set was also limited to 10 tasks with one run per task.

240

Anthropic

2026-09-28 08:40:11

Claude Code adds a wrap-up allowance so tasks can finish the current step before hitting the wall

Anthropic has changed how Claude Code behaves when users hit its 5-hour usage cap. Instead of cutting off a coding task immediately, the service now activates a feature called Wrap-Up Allowance, which lets Claude finish the step already in progress before stopping. The change addresses one of the most common complaints from developers: interrupted edits that leave repositories in a half-modified state, with function signatures changed, call sites only partly updated, or tests stopped mid-run. The allowance does not add extra usage. Anthropic says the wrap-up budget is carved out of a user’s weekly quota, applies only to the current run, cannot be saved or requested manually, and does not guarantee that the full task will be completed. Pro users can use it once per week, while Max and Team Premium users can access it each time they hit the 5-hour limit as long as weekly quota remains. The feature requires Claude Code version 2.1.277 or later, works in the terminal, IDE plugin, and desktop Code page, and is being rolled out gradually. It does not apply to API key usage, Amazon Bedrock, Google Cloud Vertex AI, or the standard Claude chat window.

290