Source-checked build studyAI & automation

Inside Hob: ask before the local agent changes a file

A source-checked look at a free local desktop agent, the boundaries it makes visible and a smaller workflow worth learning first.

INSIDE THE BUILD
The pieces behind the product
Electron desktop shellConfirmed by the public repository and release targets
llama.cpp local inferenceConfirmed repository component for running downloaded models locally
Chosen local folders and approval-before-write flowConfirmed product boundary; not a security or backup guarantee
Optional web search and MCP connectorsConfirmed network-capable extensions that require a separate trust review
Explore the documented roles belowIllustrative overview. Connections are not a verified integration diagram.

Product Hunt #13 · 99 points · 28 October 2026

CONFIRMED COMPONENTS

Inside the toolkit

4 documented components, with their roles and original sources.

Documented components, not a complete production inventory.

Electron desktop shell
Confirmed by the public repository and release targets
llama.cpp local inference
Confirmed repository component for running downloaded models locally
Chosen local folders and approval-before-write flow
Confirmed product boundary; not a security or backup guarantee
Optional web search and MCP connectors
Confirmed network-capable extensions that require a separate trust review
01

Launch evidence, with a date

Hob appeared at number 13 with 99 points on Product Hunt's daily board for 28 October 2026, checked 29 October 2026. That is evidence of launch-day attention on that board, not active-user, revenue, retention or customer-success evidence. FreeStack does not portray Hob as a newly created company or infer popularity beyond the cited ranking.

02

What the maintainers confirm

Hob's official site and public repository describe a free MIT-licensed desktop app for macOS, Windows and Linux. The application runs open-weight models locally, needs no Hob account or subscription, keeps chats and files on the machine and reports no telemetry. Users choose folders, and the product asks before file changes and supports undo. Internet use remains for downloads and optional web search or MCP connectors.

03

The production components we can actually name

The public repository identifies an Electron desktop application and llama.cpp-based local inference. The official site names compatible open-weight model families including Gemma and gpt-oss, while the chosen download determines the model licence and hardware boundary. The maintainers publish an 8 GB RAM minimum, say 16 GB is useful, and suggest 10–25 GB free disk for common models or roughly 75 GB for the largest. FreeStack does not infer undisclosed hosting, training, analytics or commercial infrastructure.

04

The useful product decision is visible permission

A folder choice, an approval before a write and an undo path make the action boundary visible to the person using the app. That does not establish enterprise access control, independent backup, malware resistance or regulatory compliance. The editorial lesson is to make every automation name the files it can read, the files it may write and the person who approves the change.

05

A smaller learning workflow

The linked FreeStack guide is an achievable exercise, not Hob's production stack: use fictional working copies, convert inspectable text with MarkItDown, ask Hob for one source-linked local brief and checkpoint the decision with Git. It teaches folder scope, review and an upgrade trigger without claiming hands-on testing or a completed customer result.

06

Where payment may become justified

Local free software can remove an app subscription while leaving hardware, support, backup and operations to the user. Pay for a specific gap such as governed sharing, central retention, supported OCR, a larger approved model, managed uptime or cross-device access. Compare that requirement with the real cost and risk of operating the local route before adding a membership.

Sources & further reading

Follow the source for current details. Product Hunt is used for discovery and launch history; technical claims use provider documentation or the project repository.

SPONSORED · THE DIGITAL ACADEMYSearch your site. See what you might need.

Start with a TDA website audit to find your next useful improvement.