Local AI Daily

Sources that discuss private, local, open weights, open source, AI

Cadence: Daily
Length: 2 minutes

Subscribe, Combine, Customize

Subscribe to this podcast
?Receive all episodes to this podcast in the apps below or anywhere that supports RSS.
Combine these episodes into your pod
?All episodes from this podcast will be fed into your own.
Sign up to add to your own podcast
Customize this pod with your own sources
?Use this if you want a brand new podcast with its own episodes using different sources.
Sign up to customize this pod

Sources

Episodes

Local AI Daily August 21: OpenCode Offers Ox Alpha With 1M-Token Context, Near-Unlimited Go Access
Created: August 21st, 2026 - 04:30 PT
Script

Here is today's Local AI Daily for Friday August 21st. OpenCode has made its new “Ox Alpha” model available to Go subscribers today, extending a promotion announced yesterday. The company describes Ox as a stealth model with a one-million-token context window, multimodal support, and zero data retention. For the next six days, OpenCode says Go users receive near-unlimited free usage that will not count against their normal allowance. The scale of the offer is notable: OpenCode says it has capacity for 100 trillion tokens a day. The big unanswered question, deliberately, is who built Ox and how it performs on real coding tasks. [1]

Also today, Pi highlighted a new academic study comparing seven software agents across five models. Its central finding: for tasks in domains with mature command-line tools, agents without built-in MCP integration completed work just as reliably while costing five to 28 times less. The researchers also found that agents would use MCP even when instructed not to if the option remained available. Pi’s takeaway is practical: for well-defined, repeatable work, a small, purpose-built harness can outperform a larger general-purpose agent setup. [2]

Yesterday, Ollama said Kimi K3 had reached more than half of its cloud subscription base for included usage, with further rollout continuing today. The service says the model is hosted in the U.S. and Europe with zero data retention. Ollama is also promising clearer performance-per-dollar pricing inside its cloud offering, and is positioning Kimi K3 as accessible through the developer tools people already use. [3]

The emerging pattern is that local and open AI is becoming less about simply downloading a model and more about choosing an operating model. Developers are weighing lean command-line workflows against broader protocol layers, using cloud capacity when it is economical, and demanding portability across tools and providers. As generous inference promotions multiply, the differentiator may increasingly be trustworthy agent behavior and predictable costs rather than raw token access.

Thank you for listening to Local AI Daily from The Daily FM. See you tomorrow!

Source Evidence
  1. Recent tweets from @opencode
    Ox Alpha is now available on OpenCode Go too
    
    For the next 6 days, usage is near unlimited and completely free
    
    It won’t count against your Go usage https://t.co/Bq4O6v3xM9
    Posted: 2026-08-21T11:11:20.000Z
    Tweet: https://x.com/opencode/status/2090758645499728234
    Links: https://twitter.com/opencode/status/2090544355824038300
    
    Ox Alpha (stealth model) is free for the next week
    
    - 1M Context
    - Multi-modal
    - Zero Data Retention
    
    Generous rate limits, near unlimited usage
    
    We have capacity for 100T tokens per day, lets see what you can do
    Posted: 2026-08-20T20:59:49.000Z
    Tweet: https://x.com/opencode/status/2090544355824038300
    
    Muse Spark 1.2 contributor from @aiatmeta is avail...
  2. Recent tweets from @pidotdev
    ...neral purpose one
    
    https://t.co/arURemqw7c
    Posted: 2026-08-21T11:30:29.000Z
    Tweet: https://x.com/pidotdev/status/2090763464553750924
    Links: https://arxiv.org/pdf/2608.08654
    
    This week we read research from a team of academics that ran a software task across 7 agents and 5 models. 
    
    They found that in domains with a mature CLI ecosystem, agents without MCP baked in completed the task just as reliably and were 5-28x cheaper. 
    
    Full arXiv paper below https://t.co/DOtkMeqpoC
    Posted: 2026-08-21T11:30:28.000Z
    Tweet: https://x.com/pidotdev/status/2090763462217551976
    Links: https://x.com/pidotdev/status/2090763462217551976/photo/1
    
    This blog post was written for those who may be curious to know what an agent harness is, but don’t, and might not know where to start. Read the full post here
    
    https://t.co/pUnUnQvEHJ
    Posted: 2026-08-20T12:30:18.000Z
    Tweet: https://x.com/pidotdev/...
  3. Recent tweets from @ollama
    ...ion base for included usage, and we're continuing to expand access today.
    
    US and Europe-hosted and zero data retention. https://t.co/eXnKlRTsrt
    Posted: 2026-08-20T18:23:33.000Z
    Tweet: https://x.com/ollama/status/2090505028998140182
    Links: https://twitter.com/ollama/status/2089914983840989255
    
    Congratulations to the whole @GoogleDeepMind Gemma team! 
    
    Gemma is one of the most popular open models. 
    
    It's been an amazing journey being a close partner! Can't wait to see what's to come! 🎉 https://t.co/CvTkgvFyt9
    Posted: 2026-08-20T18:23:27.000Z
    Tweet: https://x.com/ollama/status/2090505006785188151
    Links: https://twitter.com/osanseviero/status/2090490264112738579
    
    Model page: 
    
    https://t.co/DXMhMhBpCk
    Posted: 2026-08-19T03:18:56.000Z
    Tweet: https://x.com/ollama/status/2089914986797908051
    Links: https://ollama.com/library/kimi-k3
    
    .@Kimi_Moonshot Kimi K3 is starting to ro...
Sources
Local AI Daily August 20: OpenCode Cuts GPT-5.6 Sol Pricing as Nous Secures Hermes Skills
Created: August 20th, 2026 - 09:44 PT
Script

Here is today's Local AI Daily for Thursday August 20th. Pi is putting the spotlight on the agent harness: the layer that turns a language model into an actual working agent. In a new post published today, Earendil describes its four core pieces: a system prompt, tools, an agentic loop, and a translation layer that lets the harness work across different models. Pi says its new harness is seeing substantial work in the development branch, signaling that agent infrastructure—not just model choice—is becoming the key surface for product differentiation. [1]

Security around reusable agent capabilities is also getting more serious. Yesterday, Nous Research said Hermes now runs NVIDIA’s SkillEvaluator when users install skills. The checks look for personal data, exposed secrets, Unicode smuggling, licensing issues, and other security concerns before installation is confirmed. Nous says it tested the system on its own bundled skills and improved eleven of them based on the results. As agents gain access to files, shells, and external services, skill supply-chain security is quickly becoming a baseline requirement. [2]

On pricing, OpenCode made a big push yesterday to lower the cost of high-volume coding-agent use. Its Go subscribers can get GPT-5.6 Sol at half price through September 18th, and Hy3 users receive eight times normal usage through August 30th—advertised as 480 dollars of usage for a 10-dollar subscription. It also rolled out Muse Spark 1.2 Contributor, which offers very high limits in exchange for an explicit opt-in allowing training on user data, with regional restrictions. [3]

Finally, Nous today contrasted Hermes Cloud with premium hosted-agent subscriptions, arguing that a small idle Hermes instance can cost hundreds of times less than an idle SuperGrok subscription. Its broader pitch is flexibility: users can connect providers and models, run locally for free, or host agents remotely. [4]

The larger trend is clear: AI-agent competition is moving beyond benchmark scores. The important questions are becoming who owns the harness, how safely skills are installed, where workloads run, and whether pricing supports agents that work continuously rather than occasionally.

Thank you for listening to Local AI Daily from The Daily FM. See you tomorrow!

Source Evidence
  1. Recent tweets from @pidotdev
    This blog post was written for those who may be curious to know what an agent harness is, but don’t, and might not know where to start. Read the full post here
    
    https://t.co/pUnUnQvEHJ
    Posted: 2026-08-20T12:30:18.000Z
    Tweet: https://x.com/pidotdev/status/2090416132817670578
    Links: https://earendil.com/posts/what-is-a-harness/
    
    A harness turns a model into an agent. At it’s core it provides 4 things:
    
    - a system prompt
    - tools
    - an agentic loop 
    - a translation layer across models 
    
    New blog post from Earendil co-founder @colindaymond on what a harness is, and how you can own yours. Full post below https://t.co/8ficU3Q3xV
    Posted: 2026-08-20T12:30:00.000Z
    Tweet: https://x.com/pidotdev/...
  2. Recent tweets from @NousResearch
    ...tead of $200.
    
    https://t.co/O04Z66K1fa https://t.co/M1jwODVXLv
    Posted: 2026-08-20T13:34:47.000Z
    Tweet: https://x.com/NousResearch/status/2090432358969196548
    Links: https://portal.nousresearch.com/cloud, https://twitter.com/elonmusk/status/2090393510956401073
    
    Hermes now leverages NVIDIA's SkillEvaluator on skill installs, checking for PII, leaked secrets, Unicode smuggling, licensing and security issues before you confirm.
    
    We pointed it at our own bundled skills first, and used what it found to improve 11 of them. https://t.co/7l1cSz9UlM https://t.co/kJId0xXwFH
    Posted: 2026-08-19T19:56:53.000Z
    Tweet: https://x.com/NousResearch/status/2090166128509096187
    Links: https://x.com/NousResearch/status/2090166128509096187/photo/1, https://twitter.com/NVIDIAAI/status/2090113635683340622
    
    HY3 from @TencentHunyuan will remain free via Nous Portal through the end of the month
    
     h...
  3. Recent tweets from @opencode
    ...is restricted in some regions
    
    in exchange you get incredible usage limits https://t.co/2g76x4vnrV
    Posted: 2026-08-19T19:26:03.000Z
    Tweet: https://x.com/opencode/status/2090158371097698394
    Links: https://x.com/opencode/status/2090158371097698394/photo/1
    
    muse spark was available earlier today ahead of an official announcement
    
    capacity isn't fully ready to go so we pulled it temporarily
    
    it will be back later today
    Posted: 2026-08-19T17:39:44.000Z
    Tweet: https://x.com/opencode/status/2090131613363433526
    
    GPT-5.6 Sol is 50% off on OpenCode Zen until September 18th
    Posted: 2026-08-19T14:56:18.000Z
    Tweet: https://x.com/opencode/status/2090090487164117440
    
    Hy3 will get 8x the normal usage until August 30th
    
    that's $480 of usage for $10
    
    available to all OpenCode Go subscribers
    Posted: 2026-08-19T14:37:06.000Z
    Tweet: https://x.com/opencode/status/2090085655397216526
    
    Opera...
  4. Recent tweets from @NousResearch
    Some numbers:
    
    - A small idle Hermes Cloud instance costs 2-300x cheaper than an idle SuperGrok sub
    - For the price of a SuperGrok sub you can run 35-50 separate Hermes Cloud instances 24/7
    Posted: 2026-08-20T13:49:58.000Z
    Tweet: https://x.com/NousResearch/status/2090436180038889954
    
    Some other big differences: 
    
    - Connect any provider, sub, or model
    - Use it completely free on your own machine or with one of the several free models on Nous Portal
    - Switch between bots mode and classic sessions mode
    - Self-improving agents that learn to execute tasks cheaper
    Posted: 2026-08-20T13:35:55.000Z
    Tweet: https://x.com/NousResearch/status/2090432645528252532
    
    Hermes Agent has its own remote c...
Sources

<- Back to library