Team Nebula
Schedule a call

Nebula Labs: open-source projects from Team Nebula.

Projects we build alongside our client work and are releasing as open source. Each one says source release coming until its code is public.

See the projectsCompare them
Source release coming· nebula labs

Blurt

A voice keyboard that transcribes and rewrites without leaving the phone.

  • Transcription on device
  • No audio leaves the phone
  • Account optional, off by default
The problem
Dictation tools send your voice to someone else's server, which rules them out for anything sensitive.
Our answer
We built Blurt to run speech recognition and rewriting on the iPhone itself, with whisper.cpp and a local language model. The keyboard has no microphone access and makes no network requests.
What changed
You speak, it types, and no audio or transcript is sent anywhere, including to us.
See Blurt
Blurt · Home
Blurt for Mac, Home
Blurt, captured from the app with demo data.
Source release coming· nebula labs

HyperScale OS

One reliable workflow for outreach across channels.

  • Multi-channel by default
  • Human handoff on real replies
  • Held to the same governance bar as client work
The problem
Growth teams were copying leads, messages, and follow-ups between separate tools. Every new channel created another inbox and another place for work to disappear.
Our answer
We built a visual workflow engine that connects targeting, research, outreach, follow-up, and reply handoff across LinkedIn, email, and social channels.
What changed
The routine work keeps moving while people focus on the replies and relationships that need judgment.
See HyperScale OS
HyperScale OS · Campaigns
Campaign · workflow
Q4 operations leaders
Running · 214 prospects
LinkedIn profile view
Warm the account before any ask
Connection request
Personalized note, platform-aware
Wait for connection
Pauses until they accept
Condition: connected?
Yes → message · No → email
Send message
LinkedIn
Send email
Fallback channel
HyperScale OS, drawn from the product. Illustration, demo data.
Source release coming· nebula labs

HyperCrawl

A safe bridge between an agent and the websites people use every day.

  • Self-hosted, no third-party data pass-through
  • Exposes any site as a callable tool
  • Runs alongside our own delivery work
The problem
Useful work often lives behind logins and changing web pages. Brittle scripts break, while third-party scraping services add another place for sensitive browsing data to travel.
Our answer
We built a self-hosted web automation engine that turns approved website actions into tools an agent can call from our own infrastructure.
What changed
Agents can research and complete web tasks while the browser session and control layer stay with us.
See HyperCrawl
hypercrawl_acquire · one run
hypercrawl_acquire permits.county.example/search · needle="permit number"maxUsd 0.05
  1. 1API replay / hydration· no embedded data$0.000
  2. 2Learned tool· none saved for this site$0.000
  3. 3DOM agent· needle found · 42 rows$0.011
  4. 4Set-of-marks· not needed
  5. 5Vision· not needed
  6. 6Specialist· not needed
budget used$0.011 of $0.05
HyperCrawl, drawn from the product. Illustration, demo data.
Source release coming· nebula labs

Nebby

A careful AI code reviewer with clear limits and a human final decision.

  • Read-only, scoped access
  • Reviews every pull request in the repositories it watches
  • Same governance bar as client work
The problem
As our codebase grew, important context could be missed between reviews. A fast reviewer helped—but only if it could inspect code without quietly gaining permission to ship it.
Our answer
We built Nebby with a read-only review identity and a separate, governed path for proposed fixes. It reviews watched repositories and leaves the merge to each repository's own protections and reviewers.
What changed
Every pull request gets a consistent second look without turning an automated reviewer into an unsupervised release manager.
See Nebby
github.com · pull request
Validate tool arguments before execute #482
tm-nebbybotrequested changes
🤖 Nebby review
Needs changes: 1 blocking.
agent/src/turn.ts:95 — [HIGH] — A model can supply missing or mistyped arguments and the runner passes them straight to execute; validate them against the tool's schema first.
NEBBY_CLIENT_WORK: NO
NEBBY_VERDICT: BLOCK
Nebby — context via NebOS — changes requested: blocking findings above.
Review required
A person decides what merges. Nebby's review is advisory.
Nebby, drawn from the product. Illustration, demo data.

What guides Labs

Cutting edge, security-first.

Security-first posture

Architecture starts from data residency, least privilege and auditability, so security review is a path to ship, not a wall at the end.

Cutting-edge infrastructure

On-device models, modern agent stacks and production retrieval, the same class of capability teams see in consumer tools.

Regulated by design

Healthcare, government, energy and industrial operators need AI that respects their boundary. Labs projects assume that from the start.

Productivity without the compromise

Privacy does not mean slow. Local-first defaults and explicit cloud choice keep people moving while compliance stays intact.

Compare

Which project, for which work.

ProjectBuilt forSource
BlurtPrivate dictation on iPhoneSource release coming
HyperScale OSMulti-channel outreachSource release coming
HyperCrawlThe web as agent toolsSource release coming
NebbyReviewing the code we shipSource release coming

Talk to Team Nebula

Labs projects share DNA with our forward-deployed work. If you need agentic systems inside your boundary, bring a real workflow to the call.