Skip to content
Umbra logo

Umbra

Serves open-weight and community models from idle Apple silicon Macs through an OpenAI-compatible API, with prompts decrypted only in process memory.

From $0.03 per useMaintained

For developersFree tier
tryumbra.dev ↗Launched on Product Hunt · Jul 13, 2026
Open tryumbra.devAlternativesSet it up for my team
Umbra: data, price, alternatives
At a glancechecked Oct 31, 2026

25 linking domains

Emerging

For
Developers and Mac owners who need private inference on uncensored or community Hugging Face models.
Price
From $0.03 per usepricing page, read Oct 22, 2026
Activity
Maintainedrelease 4 months ago
How it runs
API
Used for
  • Draft confidential documents with an uncensored model
  • Point an existing OpenAI client at community models
  • Run fine-tuned Hugging Face models on Apple silicon
read off tryumbra.dev on Oct 22, 2026
Worth knowingcomputed from what we read · each with its source
  • Publishes llms.txt, a file written for AI assistantstryumbra.dev ↗
  • Launched on Product Hunt on Jul 13, 2026Product Hunt ↗
Pricingpricing page, read Oct 22, 2026

From $0.03 per use

pricing page

Price Watch: every price change in the Atlas, weekly

The companyeach fact with its source · read Oct 22, 2026
Your datawhat they publish, each line with its page · read Oct 22, 2026

These are Umbra's own statements, read from its pages. We did not audit them. A line that is missing was not found on the pages we read; it does not mean the practice is absent.

How a program reaches itfound, not claimed · read Oct 22, 2026

This card as JSON, for your agent →

Each line is something a program found when it went looking: the registry entry, the server's own answer, a spec that parses. A missing line means we did not find it, not that it does not exist.

Not it? Say what you need

beta

Three tools from the Atlas for the job you describe, each with what fits and what does not.

From people who use itnot part of the reading

No notes yet. What do you use Umbra for?

Sign in to leave a note
Questions this page answers
  • What is Umbra?

    Serves open-weight and community models from idle Apple silicon Macs through an OpenAI-compatible API, with prompts decrypted only in process memory.

  • Who is Umbra for?

    Developers and Mac owners who need private inference on uncensored or community Hugging Face models.

  • How much does Umbra cost?

    Paid plans start at $0.03 per use, per Umbra's pricing page on 2026-10-22.

  • Is Umbra free?

    No. Its pricing page showed no free plan on 2026-10-22.

  • Do people use Umbra?

    Emerging.

  • Is Umbra still maintained?

    Maintained: release 4 months ago.

  • Is Umbra safe?

    We do not audit vendors. What Umbra publishes, read 2026-10-22: a trust center.

  • Does Umbra give refunds?

    In its own words (terms of service, 2026-10-22): “Purchased credit is spendable on inference only and is non-refundable and non-withdrawable (no cash value, cannot be paid back to you).”

How we read a tool

No votes, no reviews, no vendor claims. Every week a crawler reads each tool's own site and a few public registries, and the words on the card are bands over what it read.

  • Use: visits to the site (estimated), installs from npm and PyPI, presence in Chrome's usage report.
  • Activity: the newest release on GitHub, npm or PyPI; the newest dated page on the site; open roles on a public jobs board.
  • Price: the vendor's own pricing page, read with the date. A figure is printed only when it is on that page.
  • The line and the use cases are written by us from the homepage, one tool at a time, and refused when they repeat the vendor's marketing.

How AI assistants read each site — the reading vendors ask us about — is on each tool's own visibility page.

How we read every tool → · By INITE Atlas Desk