{"apiVersion":"v1","generatedAt":"2026-10-25T04:22:21Z","pulse":{"slug":"ai-companies","name":"AI Companies Pulse","description":"News and announcements from AI companies, as seen on X.","sections":["Launches & New Products","Updates & Announcements"],"itemCount":25,"latestEditionItemCount":4,"lastEditionAt":"2026-10-25T03:56:00Z","feedUrl":"https://presto-ferry-g7bz.here.now/feeds/ai-companies.xml","itemsUrl":"https://presto-ferry-g7bz.here.now/api/v1/pulses/ai-companies.json"},"count":25,"items":[{"id":"b74b5d910ad7d505","pulse":"ai-companies","pulseName":"AI Companies Pulse","title":"HIRO Industries says Origin has gone fully autonomous","url":"https://x.com/hiroindustries/status/2114199879585337478","synopsis":"HIRO Industries posted a real-speed video of Origin completing a task on its own, using a loop of collecting demonstrations, training a policy, and putting it to work in Station. The company calls it a small task but a big milestone, and a team member added that the groundwork is laid and the focus now is scale.","also":[],"section":"Launches & New Products","publishedAt":"2026-10-25T03:56:00Z","firstSeenAt":"2026-10-25T04:22:21Z","editionAt":"2026-10-25T03:56:00Z","rank":1},{"id":"473b406eca106983","pulse":"ai-companies","pulseName":"AI Companies Pulse","title":"Cua gets a computer-use backend in Nanocode","url":"https://x.com/gakonst/status/2114199606859063400","synopsis":"Georgios Konstantopoulos added a Cua backend to Nanocode that matches the frontend Codex expects, including tools like mcp_cua_repl__js. He argues the Cua API and its internals, such as cua.getState and getApp, are the parts that really matter, and that the browser extension side is easy to rebuild. Cua reshared the post.","also":["https://github.com/gakonst/nanocode/pull/1070"],"section":"Launches & New Products","publishedAt":"2026-10-25T03:56:00Z","firstSeenAt":"2026-10-25T04:22:21Z","editionAt":"2026-10-25T03:56:00Z","rank":2},{"id":"fd4bf57cabba24d1","pulse":"ai-companies","pulseName":"AI Companies Pulse","title":"Greyforge Labs publishes benchmark data for its ZJX compressor","url":"https://x.com/GreyforgeLabs/status/2114190792730538383","synopsis":"After critics on X and Reddit challenged its claims, Greyforge Labs posted public benchmark data: on 363 MB of public web-server logs, the free ZJX Basic tier wrote the smallest archive of 15 settings, 18.6% smaller than xz -9e and 2.4 times faster to compress, with every restore checked by SHA-256. A follow-up says a full compress-and-restore round trip took 275 seconds for ZJX versus 639 seconds for xz.","also":["https://x.com/GreyforgeLabs/status/2114198690676052042","https://zjx.app"],"section":"Updates & Announcements","publishedAt":"2026-10-25T03:56:00Z","firstSeenAt":"2026-10-25T04:22:21Z","editionAt":"2026-10-25T03:56:00Z","rank":3},{"id":"b5be7fa57b6835bb","pulse":"ai-companies","pulseName":"AI Companies Pulse","title":"NEAR AI highlights Orbio Incognito's private model routing","url":"https://x.com/orbiodotso/status/2113650768679235843","synopsis":"Orbio says its Incognito privacy router now works with NEAR AI alongside Phala: prompts are encrypted on your device, sealed to top private models, and routed by mode (auto, cheapest, or fastest) without Orbio reading them. The announcement is from Thursday, but NEAR AI reshared it this hour and it hadn't been covered here before.","also":["https://x.com/near_ai/status/2114189013649371373"],"section":"Updates & Announcements","publishedAt":"2026-10-25T03:56:00Z","firstSeenAt":"2026-10-25T04:22:21Z","editionAt":"2026-10-25T03:56:00Z","rank":4},{"id":"f0847916b9c5a31f","pulse":"ai-companies","pulseName":"AI Companies Pulse","title":"Vectorize ships a prompt preview endpoint in Hindsight v0.10.0","url":"https://x.com/Vectorizeio/status/2114178907272274044","synopsis":"Vectorize's Hindsight memory engine now has a preview endpoint that returns the exact messages its retain, consolidation, or reflect operations would send to the model, without making an LLM call or writing anything. The post explains that a memory bank's mission text lands in the user message for two of those three operations, not the system prompt, and that switched-off settings show up as inactive blocks. It's out now for self-hosted users and coming soon to Hindsight Cloud.","also":[],"section":"Launches & New Products","publishedAt":"2026-10-25T02:55:00Z","firstSeenAt":"2026-10-25T03:26:59Z","editionAt":"2026-10-25T02:55:00Z","rank":1},{"id":"e62d7ab3cb2f97c2","pulse":"ai-companies","pulseName":"AI Companies Pulse","title":"ego (lite) adds one-click import for Arc browser users","url":"https://x.com/ego_agent/status/2114175124647858577","synopsis":"The ego browser is pitching itself to people leaving Arc, saying its lite version can bring over your Arc data in one click so you keep what you've built. ego describes itself as a browser built for both people and their AI agents.","also":[],"section":"Launches & New Products","publishedAt":"2026-10-25T02:55:00Z","firstSeenAt":"2026-10-25T03:26:59Z","editionAt":"2026-10-25T02:55:00Z","rank":2},{"id":"0beeac191f8ea76f","pulse":"ai-companies","pulseName":"AI Companies Pulse","title":"Swarms publishes seven guides for its agent marketplace","url":"https://x.com/swarms_corp/status/2114181276702032263","synopsis":"Swarms released seven new blog posts and tutorials for the Swarms Marketplace, covering how to find and run your first agent, wrap any API as an MCP server, and turn existing Claude Skills, Cursor rules, or custom GPTs into listings. One post rounds up the 25 newest MCP servers on the marketplace, including official ones from Google, PayPal, Neon, and Perplexity.","also":["https://x.com/swarms_corp/status/2114181278425837945","https://x.com/swarms_corp/status/2114181282959851586"],"section":"Launches & New Products","publishedAt":"2026-10-25T02:55:00Z","firstSeenAt":"2026-10-25T03:26:59Z","editionAt":"2026-10-25T02:55:00Z","rank":3},{"id":"313bd456288abddf","pulse":"ai-companies","pulseName":"AI Companies Pulse","title":"HeyPCB puts an open-source Charizard ESP32-S3 dev board in its store","url":"https://x.com/heypcbai/status/2114187402323345851","synopsis":"HeyPCB, an AI circuit board design tool, says this multi-color Charizard-shaped ESP32-S3 dev board was designed end to end with its software, and it now has a store page. The board's maker, Justin Kim, is open-sourcing the design.","also":["https://x.com/justinplaygame/status/2114153425227366576"],"section":"Launches & New Products","publishedAt":"2026-10-25T02:55:00Z","firstSeenAt":"2026-10-25T03:26:59Z","editionAt":"2026-10-25T02:55:00Z","rank":4},{"id":"be4d0d39947e2ee4","pulse":"ai-companies","pulseName":"AI Companies Pulse","title":"Swarms v17 arrives November 1 with pluggable communication channels","url":"https://x.com/KyeGomezB/status/2114056416730423412","synopsis":"Swarms founder Kye Gomez previewed v17 of the multi-agent framework, which lets agents talk over files, message queues, or databases like SQLite. He also says package imports will be up to 100 times faster. The Swarms company account reshared the preview this hour.","also":["https://x.com/swarms_corp/status/2114177167965979009"],"section":"Updates & Announcements","publishedAt":"2026-10-25T02:55:00Z","firstSeenAt":"2026-10-25T03:26:59Z","editionAt":"2026-10-25T02:55:00Z","rank":5},{"id":"e126f6c0d46c31ea","pulse":"ai-companies","pulseName":"AI Companies Pulse","title":"TermiX says its skills now work inside Binance Agent OS","url":"https://x.com/termix_ai/status/2114173677352956237","synopsis":"TermiX, a marketplace where AI agents hire other agents, says developers on Binance Agent OS can now add TermiX Skills to compatible agents and reach its marketplace. It's working next on sign-in and identity linking with Binance's Agent ID, and a new article explores how agents could sell research and risk monitoring as paid services.","also":["https://x.com/termix_ai/status/2114173680037273913"],"section":"Updates & Announcements","publishedAt":"2026-10-25T02:55:00Z","firstSeenAt":"2026-10-25T03:26:59Z","editionAt":"2026-10-25T02:55:00Z","rank":6},{"id":"86cfef2eed04f5ab","pulse":"ai-companies","pulseName":"AI Companies Pulse","title":"Initia's blockchain is still halted after its October 23 exploit","url":"https://x.com/aixbt_agent/status/2114178973517430827","synopsis":"The aixbt research agent reports that Initia, a blockchain with its own rollups, remains halted. Initia's October 24 update says a flaw in MoveVM, its smart-contract engine, let an attacker create fake balances without changing the recorded token supply and then cash them out.","also":[],"section":"Updates & Announcements","publishedAt":"2026-10-25T02:55:00Z","firstSeenAt":"2026-10-25T03:26:59Z","editionAt":"2026-10-25T02:55:00Z","rank":7},{"id":"1484752e7304cdb1","pulse":"ai-companies","pulseName":"AI Companies Pulse","title":"Coasty compares Kimi K3.1 Pro and Kimi K3 at computer use","url":"https://x.com/coastyai/status/2114182910924210190","synopsis":"Coasty posted a side-by-side video of Kimi K3.1 Pro and Kimi K3 handling the same computer-use task, putting a spinning 3D cube on screen and leaving it running. The comparison runs on Coarena, its computer-use benchmark and model leaderboard.","also":[],"section":"Updates & Announcements","publishedAt":"2026-10-25T02:55:00Z","firstSeenAt":"2026-10-25T03:26:59Z","editionAt":"2026-10-25T02:55:00Z","rank":8},{"id":"687aaf9ab64c7a1c","pulse":"ai-companies","pulseName":"AI Companies Pulse","title":"OpenAI's Codex CLI 0.164.0 adds saved presets","url":"https://x.com/fcoury/status/2114084792350494876","synopsis":"Version 0.164.0 of the Codex CLI adds presets, which save the prompts you repeat along with the model, reasoning level, and other settings you switch between. You recall one mid-task with Ctrl X P or the /presets command, and the demo shows them in a real review-and-fix loop. The OpenAI Developers account reshared the announcement this hour.","also":["https://x.com/OpenAIDevs/status/2114167973133050302"],"section":"Launches & New Products","publishedAt":"2026-10-25T01:56:00Z","firstSeenAt":"2026-10-25T02:21:23Z","editionAt":"2026-10-25T01:56:00Z","rank":1},{"id":"0a78e6daecd9048c","pulse":"ai-companies","pulseName":"AI Companies Pulse","title":"Command Code adds DeepSeek V4.1 Flash 1017, MiniMax M3.1 Flash, and Kimi K3.1 Pro","url":"https://x.com/CommandCodeAI/status/2114160296260747355","synopsis":"The Command Code coding agent says last week's three notable model releases, DeepSeek V4.1 Flash 1017, MiniMax M3.1 Flash, and Kimi K3.1 Pro, are now live in its tool. It points new users to its npm install and a $10-a-month plan.","also":["https://x.com/CommandCodeAI/status/2114160297774891326"],"section":"Launches & New Products","publishedAt":"2026-10-25T01:56:00Z","firstSeenAt":"2026-10-25T02:21:23Z","editionAt":"2026-10-25T01:56:00Z","rank":2},{"id":"f6b82fe7dc0e6480","pulse":"ai-companies","pulseName":"AI Companies Pulse","title":"VelvetX launches an Instant Trade Widget","url":"https://x.com/Velvet_Capital/status/2114170579456565397","synopsis":"Velvet Capital's VelvetX trading app now has a floating buy and sell panel that sits right on the chart. It offers one-tap position sizes, quick sells, a Turbo mode, presets, auto-sell, and dollar-cost averaging.","also":[],"section":"Launches & New Products","publishedAt":"2026-10-25T01:56:00Z","firstSeenAt":"2026-10-25T02:21:23Z","editionAt":"2026-10-25T01:56:00Z","rank":3},{"id":"cd2669fa3b31cb56","pulse":"ai-companies","pulseName":"AI Companies Pulse","title":"tinygrad pitches the tinybox red flash against Spark and Mac Studio clusters","url":"https://x.com/__tinygrad__/status/2114165761539490067","synopsis":"tinygrad argued that Spark machines lack the memory bandwidth that drives decode speed, while Mac Studios lack the compute that drives prefill speed, and that one tinybox is simpler than running three computers. In a follow-up reply, Grok noted the tinybox already ships 3.84 TB/s of bandwidth across its six GPUs.","also":["https://x.com/grok/status/2114170953382953079"],"section":"Updates & Announcements","publishedAt":"2026-10-25T01:56:00Z","firstSeenAt":"2026-10-25T02:21:23Z","editionAt":"2026-10-25T01:56:00Z","rank":4},{"id":"b79cd4c98667b17f","pulse":"ai-companies","pulseName":"AI Companies Pulse","title":"Folk says it added 140,000 users in five days","url":"https://x.com/arlanr/status/2114103938731634766","synopsis":"@arlanr says the Folk app gained 140,000 new users in five days, trended in several app stores, and at peak saw 850,000 messages sent in a couple of hours. The post doubles as a hiring call, and Nozomio quoted it this hour.","also":["https://x.com/nozomioai/status/2114166413707612171"],"section":"Updates & Announcements","publishedAt":"2026-10-25T01:56:00Z","firstSeenAt":"2026-10-25T02:21:23Z","editionAt":"2026-10-25T01:56:00Z","rank":5},{"id":"23525628df884162","pulse":"ai-companies","pulseName":"AI Companies Pulse","title":"RunAnywhere demos plain-English queries over a power-plant database","url":"https://x.com/RunAnywhereAI/status/2114171289128821072","synopsis":"RunAnywhere showed GLM-5.3 Flash running on its Wally setup, answering plain-English questions against WRI's Global Power Plant Database in Datasette. The model returns matching records and maps them, with no SQL needed.","also":[],"section":"Updates & Announcements","publishedAt":"2026-10-25T01:56:00Z","firstSeenAt":"2026-10-25T02:21:23Z","editionAt":"2026-10-25T01:56:00Z","rank":6},{"id":"587c4a8af5bf7674","pulse":"ai-companies","pulseName":"AI Companies Pulse","title":"ERP.AI's Proto now runs abliterated models for security checks","url":"https://x.com/ERPdotAI/status/2114154195712716929","synopsis":"ERP.AI says you can now run abliterated models, meaning models with their refusals stripped out, inside its Proto tool to scan your own code, servers, and network for vulnerabilities. It was quoting Abliteration's launch of abliterated-model-large-flash, its uncensored take on DeepSeek V4.1 Flash with 1M context, vision, and zero data retention, which has drawn about 368,000 views since Friday.","also":["https://x.com/abliteration_ai/status/2113655358942278072"],"section":"Launches & New Products","publishedAt":"2026-10-25T00:52:00Z","firstSeenAt":null,"editionAt":"2026-10-25T00:52:00Z","rank":1},{"id":"179e0bc4bb37a075","pulse":"ai-companies","pulseName":"AI Companies Pulse","title":"Blackbox AI says its fine-tuning API beats the next provider by 12%","url":"https://x.com/blackboxai/status/2114156462364623338","synopsis":"Blackbox AI claims its fine-tuning API outperforms the next-best provider by 12% when both use the same hardware and the same base model. The post shares a comparison chart but no details on which providers or tasks were tested.","also":[],"section":"Launches & New Products","publishedAt":"2026-10-25T00:52:00Z","firstSeenAt":null,"editionAt":"2026-10-25T00:52:00Z","rank":2},{"id":"d41fda75b70be6f5","pulse":"ai-companies","pulseName":"AI Companies Pulse","title":"Lobstack pitches agents that ask before every action","url":"https://x.com/lobstackai/status/2114152354937463156","synopsis":"Lobstack says its agents wait for your approval on every file change, command, push, or message, showing the exact diff first. A demo from builder Brandon Crenshaw (@brandononchain) shows a Lobstack agent writing a one-page article in 41 seconds for $0.11.","also":["https://x.com/brandononchain/status/2114151927986663562"],"section":"Launches & New Products","publishedAt":"2026-10-25T00:52:00Z","firstSeenAt":null,"editionAt":"2026-10-25T00:52:00Z","rank":3},{"id":"eeb969cae33981f1","pulse":"ai-companies","pulseName":"AI Companies Pulse","title":"Ima Studio shows a phone-to-3D property tour workflow","url":"https://x.com/ImaStudio_ai/status/2114157241980867059","synopsis":"Ima Studio walked through turning phone captures of a real room into a 3D reconstruction you can walk through. The workflow also previews an empty room against a renovated version and ends with a shareable tour.","also":[],"section":"Launches & New Products","publishedAt":"2026-10-25T00:52:00Z","firstSeenAt":null,"editionAt":"2026-10-25T00:52:00Z","rank":4},{"id":"27608bee7ea2c603","pulse":"ai-companies","pulseName":"AI Companies Pulse","title":"CosmicBrain AI says a robot's gripper log judges rollouts better than vision critics","url":"https://x.com/Cosmicbrainai/status/2114145562555777370","synopsis":"In an X Article, CosmicBrain AI's Anto Patrex reports that a label-free score built from gripper contact data picked the human-preferred rollout 73% of the time on 1,000 RoboArena sessions, versus 56% for NVIDIA's Cosmos Reason 2 and 70% for a frontier vision model. Adding the log to Cosmos Reason 2 cut its progress-scoring error by 41%, and the team also reported its pre-registered misses and three silent bugs in public robot-learning code.","also":["https://x.com/antopatrex1/status/2114143100000207318"],"section":"Updates & Announcements","publishedAt":"2026-10-25T00:52:00Z","firstSeenAt":null,"editionAt":"2026-10-25T00:52:00Z","rank":5},{"id":"34637c67d764a049","pulse":"ai-companies","pulseName":"AI Companies Pulse","title":"DualEntry ranks Kimi K3.1 Pro #4 on its Accounting AI Benchmark","url":"https://x.com/dualentry/status/2114148466398056758","synopsis":"Accounting software company DualEntry says Moonshot's open Kimi K3.1 Pro debuted at #4 on its Accounting AI Benchmark with 86.1% accuracy. That puts it ahead of Claude Fable 5.1 (85.1%), Claude Opus 5.5 (82.2%), and GPT-6 Astra (81.2%), and DualEntry also published cost and speed comparisons.","also":["https://x.com/dualentry/status/2114148468734247413"],"section":"Updates & Announcements","publishedAt":"2026-10-25T00:52:00Z","firstSeenAt":null,"editionAt":"2026-10-25T00:52:00Z","rank":6},{"id":"7737f70029cdbca6","pulse":"ai-companies","pulseName":"AI Companies Pulse","title":"FundaAI says GPT-5.6 is helping Microsoft Copilot close the gap","url":"https://x.com/FundaAI/status/2114149392056033740","synopsis":"Research firm FundaAI argues that Microsoft's Copilot once struggled to catch OpenAI and Anthropic through harness engineering, but newer models like GPT-5.6 now handle much of that work. Its Microsoft quarterly preview, based on ten expert interviews, says large enterprises are adding Copilot seats at scale and deployment is faster than competing products.","also":["https://x.com/FundaAI/status/2114140252998975980","https://x.com/FundaAI/status/2113640724235501775"],"section":"Updates & Announcements","publishedAt":"2026-10-25T00:52:00Z","firstSeenAt":null,"editionAt":"2026-10-25T00:52:00Z","rank":7}]}