The MindsHub press kit.
Everything you need to write about, link to, or design alongside MindsHub — every major model behind one endpoint, plus MindsHub Cowork, the agent workspace on top of it, by the team behind MindsDB.
- 12 Logo variants
- SVG Vector files
- 4 Color ramps
- 2 Typefaces
01 · About MindsHub
Where open-source models get things done.
In 2018, two founders in Berkeley asked a simple question: why does using AI mean copying your data somewhere else? They open-sourced an answer — bring the models to where the data already lives — and the principle held through 500,000+ deployments and a community of builders that grew up around it.
MindsHub is the next chapter of that work. As the field moved from one model call to many — with tools, memory, and execution at every step — that principle needed both an engine and a workspace. That's MindsHub: Unified Inference puts every major model behind one endpoint, and Cowork is where you delegate entire projects and collect finished, shareable results, with work running on two interchangeable open-source agent harnesses — Anton and Hermes. Same mission, wider surface. By MindsDB.
- Founded
- 2018 · Berkeley, CA
- Open source
- github.com/mindsdb ↗
- Stars
- 39,000+ on GitHub
- Pronounce
- /ˈmaɪndz hʌb/ — minds-hub
- Press email
- media@mindsdb.com
02 · Logos & lockups
The MindsHub mark, in every flavor you'll need.
The primary lockup is the bear with the MindsHub wordmark. Use the full-color version on neutral light backgrounds. Drop in a monochrome or reversed variant for dark, branded, or single-color contexts. Files ship with built-in clear space.
02.1 · Construction
How to handle the mark.
Three rules: give it room, keep it large enough to read, and don't dress it up.
Clear space
Maintain a clear zone equal to the height of the bear icon
(×) on every side. Type, imagery, and frame
edges stay outside.
Minimum size
The horizontal lockup must never render smaller than 96 px wide on screen or 20 mm wide in print. The bear-only icon can scale down to 24 px.
96 px · digital min
140 px · comfortable
Don't.
Don't stretch or squash. Keep the original aspect ratio.
Don't rotate the mark. Always set it on the horizontal baseline.
Don't re-color the mark. Use the approved palette swatches only.
Don't place on low contrast. The bear must remain clearly readable.
Don't lay over busy imagery. Use the dark variant in a flat panel instead.
Don't add effects. No shadows, glows, outlines, or strokes.
03 · Color palette
Aqua-led, on a neutral ground.
MindsHub and the Cowork workspace render from one palette, so the site and the product look like the same company. Brand aqua carries the identity; a neutral ink-and-surface ramp does the reading; frost supplies the dark bands. Click any swatch to copy its hex.
Aqua — brand
--primary-* · --aqua-* Ink — text
--ink · --ink-2…5 Surface & line — the page
--bg · --surface-* · --line-* Frost — neutrals & dark bands
--frost-* Accent — roles
--accent-* Signal — status
functional only 04 · Typography
Two typefaces, three jobs.
Inter does display and body both — one variable file, optically sized, so a headline and a paragraph get different letterforms out of the same outlines. JetBrains Mono handles code, identifiers, and the metadata that lives around the brand.
Inter
Display · UI · Body Open agents,
open by default.
One variable file covers display and body alike. Inter carries a weight axis (100–900) and an optical-size axis, so it draws display letterforms in a headline and text letterforms in a paragraph without a second file — headlines at 600/700, body at 400/500.
Variable 100–900 · upright + italic · optical sizing auto
rsms.me/inter ↗JetBrains Mono
Mono · Code · Metadata agent.run({ model: "latest:sonnet" })
Code blocks, model IDs, file paths, keyboard hints, and the quiet eyebrow labels sprinkled through the page. Shipped at Regular and Medium only — there is deliberately no bold monospace face, because bold mono has no job in this material.
500 Medium · 400 Regular · no bold face
jetbrains.com/lp/mono ↗
Both families are open source under the SIL Open Font License 1.1, and
the site serves them itself from /fonts/ as
WOFF2 rather than pulling them from a font CDN. Set your own materials
in the same two families and they will match ours exactly.
05 · Voice, tone & boilerplate
Voice & tone — plain, technical, slightly understated.
This is the MindsHub tone of voice — the way good engineering sounds: precise, low-drama, and unimpressed by hype. We name things accurately. We let the work do the lifting.
Precise > clever.
If we have to choose between a clever line and a true one, pick the true one. The audience is technical; they can tell the difference.
Is Run open-source agents with real tool access, memory, and credentials.
Not Unleash the magic of agentic AI.
Concrete > abstract.
Prefer verbs that describe what actually happens — connect, route, host, query, infer — over abstractions like "leverage" or "synergize."
Is Connect Postgres. Route inference. Persist memory across runs.
Not Unlock enterprise-grade AI capabilities at scale.
Open > closed.
MindsHub is community-first by construction. The agents are open source. The roadmap is public. Lead with what's available now, not what's gated behind a sales call.
Is Fork an agent. Read the merge thread.
Not Contact us to learn more.
One-liner · use as-is
MindsHub puts every major model behind one endpoint — one API key, one bill — and Cowork, the agent workspace on top of it, turns a briefed project into finished, shareable work.
Boilerplate · ~75 words
MindsHub puts every major model behind one endpoint. Unified Inference, the MindsHub API at api.mindshub.ai, serves frontier and open models through the request formats developers already use — OpenAI Chat Completions and Anthropic Messages — on one API key, one prepaid balance and one usage summary, with stable model aliases that follow upgrades without code changes. Cowork, the agent workspace on top of that catalog, is for the work you would rather delegate than integrate: knowledge workers brief entire projects — apps, websites, research, analysis, reporting, scheduled operations — and collect finished, shareable results. Data connects through a secure credentials vault; work runs on two interchangeable open-source agent harnesses, Anton and Hermes, with artifacts, persistent memory, skills and scheduling built in. Open by design: swap the model or the harness and keep everything you’ve built. By MindsDB. Founded 2018 in Berkeley. Backed by Benchmark, Mayfield, Y Combinator, and NVIDIA.