{"slug": "june-bundles-a-private-mac-agent-dictation-and-meeting-notes-in-one-app", "title": "June bundles a private Mac agent, dictation and meeting notes in one app", "summary": "OpenSoftware released June on June 30th as a Mac workspace that combines AI chat, voice dictation, meeting notes and a local agent, keeping application data on-device by default while routing model inference to remote providers unless users select local models. The open-source app, published under the MIT license, uses a confidential-compute backend on Phala Cloud and defaults to Venice for model calls, with privacy tiers labeled in the model picker.", "body_md": "[Gauthier Le Meur (@0xgaut)](https://x.com/0xgaut), an Alongside co-founder, released OpenSoftware's June on June 30th as a Mac workspace that combines AI chat, voice dictation, meeting notes and a local agent. OpenSoftware renewed the privacy pitch [in a July 22nd post on X](https://x.com/OpenSoftwareCo/status/2079986143706161531), emphasizing support for local models and private models supplied by Venice.\n\nThe product targets a basic conflict in desktop agents: the most useful assistants need access to files, recordings, calendars and long-lived context, which also makes them unusually invasive cloud applications. [June](https://www.opensoftware.co/june) keeps its application database, recordings, transcripts, files, sessions and agent memory on the user's Mac by default. Its agent runs locally using the open-source Hermes framework, with sessions placed in a macOS write jail unless the user opts out.\n\nThat local-first design does not mean every operation stays offline. Dictation cleanup, transcription, note generation and other model-powered features can send prompts, audio, files or context away from the device for processing. OpenSoftware's [privacy policy](https://www.opensoftware.co/legal/privacy) says those requests pass through a confidential-compute backend designed to hide their contents from OpenSoftware while they are in flight.\n\nThe distinction matters because \"local AI\" has become a loose marketing category. June stores its working state locally, but its default model inference is remote. Users can select supported local text models when they need on-device inference, a capability OpenSoftware added on July 2nd. The application also lets users bring their own Venice API key.\n\n### Open source as the privacy proof\n\nOpenSoftware has published [June's source code on GitHub](https://github.com/open-software-network/os-june) under the MIT license. The repository had 325 stars and 39 forks as of July 22nd.\n\nThe public code covers the desktop application and June API. OpenSoftware says the production backend runs in an Intel TDX confidential virtual machine on Phala Cloud, with a published container image, source revision and remote attestation report. That setup gives users a way to check which software is running inside the trusted execution environment rather than relying solely on a privacy policy.\n\nJune routes model calls to [Venice](https://venice.ai/privacy) by default. Venice divides models into several privacy classes. Its \"Private\" tier relies on Venice-controlled GPUs or providers bound to zero-data-retention terms, while its TEE and end-to-end encrypted tiers add hardware-enforced protections. OpenSoftware says anonymized third-party models are opt-in and warns in its repository that those providers may retain the content they receive under their own policies.\n\nThat caveat puts the burden on the model picker. A user can keep application state on a Mac and still send selected content to an outside provider. June labels models by privacy tier, an important product choice for an application that may handle meeting audio, email, financial files or health information.\n\nGoogle connections follow a similar split. June can store Google access tokens in the Mac's Keychain and make Gmail and Calendar requests directly from the device. When a user asks a model to process selected Google content, that content may still go to the chosen inference provider. Selecting a local model keeps that step on the Mac.\n\n### A second act for the Alongside founders\n\nJune is operated by Alongside Finance, according to OpenSoftware's [terms of service](https://www.opensoftware.co/legal/terms). Le Meur previously co-founded Alongside with Austin Diamond, Jun Ho Hong and Rai Sur. Alongside raised an $11 million seed round led by a16z crypto in 2023, with Coinbase Ventures, Franklin Templeton Investments, Village Global and Not Boring Capital participating, according to [TechCrunch](https://techcrunch.com/2023/02/15/crypto-index-platform-alongside-raises-11m-led-by-a16z/).\n\nOpenSoftware's website displays a16z crypto as a backer, but it does not disclose a separate financing round or valuation for OpenSoftware or June. The legal continuity is clearer than the financial structure: the same Delaware corporation that built Alongside's crypto products now provides the OpenSoftware network and June.\n\nLe Meur's move into private desktop AI carries over a familiar infrastructure thesis from his earlier companies. The product uses verifiable software, confidential compute and external networks to reduce the amount of trust placed in a central operator. June packages those components into a consumer-facing Mac application rather than asking users to assemble local models, agent frameworks and inference providers themselves.\n\n### June is shipping quickly, with limits\n\nOpenSoftware has released frequent updates since June's debut. Version 0.0.35, published [on July 21st](https://www.opensoftware.co/june/changelog), added isolated profiles, Notion and Obsidian connections, Google Calendar context for meeting notes, startup controls and better transcription recovery. Earlier July releases added encrypted browser sharing, local Google connectors, image generation and support for local text models.\n\nMac remains the primary platform. The repository describes Windows builds as incomplete, lacking global dictation paste, system-audio capture and the macOS sandbox.\n\nJune has a free tier with limited usage, local-model support, starter credits and bring-your-own-key access. Pro costs $20 per month and adds heavier daily use, scheduled routines and cloud models. Max costs $100 per month for higher limits and more demanding agent sessions.\n\nOpenSoftware is competing on architecture rather than a proprietary foundation model. June's bet is that users will accept a mix of local storage and remote inference when the routing layer is inspectable, model privacy is labeled and local inference remains available. The difficult part will be making those distinctions clear enough that users understand exactly when sensitive data stays on the Mac and when it leaves.", "url": "https://wpnews.pro/news/june-bundles-a-private-mac-agent-dictation-and-meeting-notes-in-one-app", "canonical_source": "https://runtimewire.com/article/june-bundles-a-private-mac-agent-dictation-and-meeting-notes-in-one-app", "published_at": "2026-07-22 19:56:52+00:00", "updated_at": "2026-07-22 20:08:42.525149+00:00", "lang": "en", "topics": ["ai-products", "ai-agents"], "entities": ["OpenSoftware", "June", "Gauthier Le Meur", "Venice", "Phala Cloud", "Alongside Finance", "Intel TDX", "Hermes"], "alternates": {"html": "https://wpnews.pro/news/june-bundles-a-private-mac-agent-dictation-and-meeting-notes-in-one-app", "markdown": "https://wpnews.pro/news/june-bundles-a-private-mac-agent-dictation-and-meeting-notes-in-one-app.md", "text": "https://wpnews.pro/news/june-bundles-a-private-mac-agent-dictation-and-meeting-notes-in-one-app.txt", "jsonld": "https://wpnews.pro/news/june-bundles-a-private-mac-agent-dictation-and-meeting-notes-in-one-app.jsonld"}}