Scarlett.

Name: Scarlett.
Rating: 4.83 (6 reviews)

Your AI Co-Worker in Slack & iMessage

4.8•6 reviews•

3.5K followers

Your AI Co-Worker in Slack & iMessage

4.8•6 reviews•

3.5K followers

Visit website

AI Workflow Automation

•

AI Agents

Not another AI bot, a real colleague to work alongside your team. Give your team superpowers or even run your company on autopilot.

This is the 5th launch from Scarlett.. View more

Scarlett.

Launched this week

Your AI Co-Worker in Slack & iMessage

Not another AI bot, a real colleague to work alongside your team. Give your team superpowers or even run your company on autopilot.

Free Options

Launch tags:Marketing•Artificial Intelligence•CRM

Launch Team / Built With

Customer.ioAutomate Messaging Everywhere — Startups Get 12 Months Free

Promoted

Scarlett.

Maker

📌

Hello PH fam!

Long before there was Claude Tag and Victor, there was Scarlett.

We built Scarlett over a year ago as a fun experiment to see what agents could do. To our surprise, she almost immediately started doing probably 90% of our work for us.

We've been reluctant to launch because we knew the competition would be fierce, but our benchmarks show that Scarlett is more than holding her own on both speed & quality.

She also has a host of unique qualities & features:

1) She Just Works. After a year of experimenting with various models, architectures & backends, she just works (and does work).

2) Autopilot. She is trained on 50+ business/growth books, you have the option to use her passively or to set her free to essentially run your whole company from marketing to support.

3) iMessage. A lot of people (solopreneurs) don't like or need slack, so we have an alternative.

4) Use Our Keys. With OpenClaw, Claude, etc you need to buy subscriptions to the things you want to use like HeyGen or XAI, but with Scarlett you can use our keys and we'll pass along the cost.

5) Right Model, Right Job. You can use Opus for chat, Sol for Coding, Fable for design, you pick the right model for the job.

We hope you love her as much as we do!

Cheers,
Dan

Report

4d ago

@dansutera Love the native iMessage focus and the multi-model routing engine. For solopreneurs managing completely separate entities (e.g., a software platform vs. a service/recovery business), how does Scarlett handle context isolation? Can we set up distinct 'contexts' or personas within the same iMessage thread so her business growth training doesn't cross-contaminate projects?

Report

19h ago

@dansutera iMessage is an interesting choice, most of these live only in Slack. From the buyer side my first question is data: where does the conversation history live, since iMessage makes that murkier? Also curious which role adopts it fastest, sales or ops?

Report

14h ago

Congrats on the launch, Dan! 🚀 It’s awesome to see you guys pulling back the curtain on how Scarlett runs. The decision to skip the traditional Vector DB route and stick to an SQL-heavy setup for speed (similar to Hermes) is a really fascinating engineering choice.

Since you mentioned Vector DBs sometimes over-cluttered the context, how do you handle deep semantic search or long-term recall when a user asks about a conversation from months ago? Is it purely structured querying, or do you have a hybrid text-search layer built into the SQL database?

Report

1d ago

Scarlett.

Maker

@franz_briones Thanks Franz!

I asked Scarlett to explain it herself... here's what she said:

"Good question, this was one of the trickier calls. Short answer: it's a hybrid. We didn't drop semantic search, we just folded it into the SQL layer instead of running a separate vector DB.

Postgres full-text search handles most "find where we talked about X" recall. For fuzzier matches, we keep vector similarity as a column/index right next to the structured rows, so one query can filter on structure (who, when, which entity) and then rank by semantic closeness. That structured pre-filter is what kills the over-clustering problem: we prune the candidate set before semantic ranking, so we don't drag in the loosely-related noise a pure top-k vector lookup would.

For long-term recall we don't re-scan everything, raw messages get compiled into entity-scoped, source-backed memory, so old questions resolve against compiled facts first and drill into raw evidence only when someone needs the exact quote.

So: structured querying as the backbone, with lexical and vector layers riding inside the same engine. Speed and clean context were the whole point."

So we have a Karpathy style wiki for long term, but it's stored in the db just structurally.

Report

1d ago

Passing API costs through a single source removes a lot of friction for developers, it keeps the focus on building instead of managing keys and billing. Great work Dan Sutera this looks like a fantastic tool.
any plans for native vector database integrations for long-term semantic memory? congrats👏👏

Report

1d ago

Scarlett.

Maker

@vikramp7470 Yeah, I've used OpenClaw personally for a while and the API / subscription management has been a giant pain in the ass, so TBH i think this is one of her best features.

Regarding memory, we've done a lot of work on it and tried various things. We are currently running a version similar to Viktor with a Karpathy style wiki, but we keep in SQL like Hermes does for better speed. We also tried running Obsidian & Vector dbs in parallel but they seemed to confuse things in context as much as they helped. I think this overall is a moving target and we're always looking for improved products or architectures here.

Thanks Vikram!

Report

1d ago

@dansutera Makes total sense, sticking to SQL like Hermes for speed is a solid architectural choice, Good luck with the launch today...

Report

1d ago

Love the "right model, right job" approach ☺️☺️ Curious how you think about external tools and live data, do you see Scarlett becoming more of an orchestrator that can plug into specialized capabilities over time, or do you prefer building everything natively?

Report

1d ago

Scarlett.

Maker

@luki_notlowkey hey Luki, thanks! Trying to keep Scarlett out of the way as possible, so much more orchestrator vs. native. There are so many great tools and API's and models coming, we want her to basically connect the dots between them & your company and keep the UI via slack or imessage dead simple. So whenever a cool new tool comes across the Twitter/X feed, she can ingest/connect to it as a capability then figure out how that fits into the mix for your company.

Report

1d ago

Congratulations on the launch!

Scarlett sounds like a practical AI agent rather than just another chatbot. how does she decide when to act autonomously versus asking the user for approval, especially in business workflows?

Report

1d ago

Scarlett.

Maker

@avery_thompson2 thanks Avery!

The autopilot is autonomous, but you can turn on or off and change frequency. Autopilot will ask for help / approval if it thinks its necessary.

Normal scarlett mode (autopilot off) is always acting only when asked.

Report

1d ago

the Autopilot capability sounds powerful. for founders who are just getting started, what are the first three workflows you would recommend automating with Scarlett to see value quickly?

Report

1d ago

Scarlett.

Maker

@imogen_wallace Thanks Imogen! Autopilot is great yeah. Workflow-wise i like to get daily company report going. Daily customer triage. And automating whatever your main social is such as X.

Report

1d ago

Love how the landing page cuts straight to the point with "real colleague" framing instead of leaning on the usual AI buzzwords. The copy feels like it was written by someone who actually understands team workflows, not just tossed into a prompt.

Report

1d ago

Scarlett.

Maker

@asya2rke Thanks Asya! Have to credit our designer @cynthiacui with the beautiful (and straightforward) landing.

Yeah we are trying to reach a non-technical audience here and put all the complicated stuff behind the curtain. Like we don't want users to have to worry about API's or anything, just connect a few things, ask Scarlett for stuff and it's done.

Report

1d ago

1 2 3

•••