Pull up a desk.
Your AI coworker does the rest.
Smalt is one AI coworker with the tools to finish real work. Ask for a DCF from a data room, a deck from a memo, a research note with sources, or a batch of outreach emails. It comes back as a file you can open, check, and send.
Built alongside the people who use it
- Investment analysts
- Private-equity associates
- Corp-dev teams
- IB analysts and associates
- Equity research
What it does
The jobs it takes off your desk.
There is no agent picker and no mode to choose. You describe the job and it uses whichever tools the job needs.
- Financial models
- DCF, LBO, three-statement, comps. Built in Excel with real formulas, so when you change an assumption the model recalculates. Sensitivity tables included when you ask for them.
- Research with sources
- Market maps, competitor scans, first-pass due diligence. Every claim links back to the filing, article, or transcript it came from, at paragraph level.
- Decks
- Native PowerPoint, editable, on your template. Insight titles, real charts, and an analyst note on each slide. Not screenshots of a chart.
- Emails and contacts
- Draft outreach, write up a meeting brief, find the right person at a company. Nothing goes out until you have read it and hit send.
- Spreadsheets and documents
- Upload a workbook or a folder and ask questions about it. Or start from nothing in Google Docs, Sheets, Word, or Excel.
- Your data stays yours
- Hosted on AWS, encrypted in transit and at rest, never used to train models. The sub-processor list is public. The details are on the Trust pages.
One job, start to finish
What a Tuesday afternoon looks like.
An associate has a board call at three and a data room with 47 files in it. She writes one line to Smalt: “Can you build the Alder DCF from the data room? Board call at three.”
Smalt reads the 47 files. It builds a 14-sheet model with the assumptions, revenue build, three statements, WACC, and valuation all linked, then runs a five-by-five sensitivity on WACC against growth. It shows the steps as it goes, so she can see what it read and what it assumed.
Twenty minutes later there is an Excel file on her desk with live formulas, and a note that one assumption in the management case looks optimistic. She opens the model, checks the flagged cell, adjusts it, and is ready for the call.
That is the whole product. No prompt engineering, no agents to configure, nothing to learn beyond describing the job the way you would to a colleague.
Product and services
Two ways to work with us.
Use the product
Sign up and start working. You get 500 credits free, no card, and the full toolbox from the first message. Most people on Smalt are analysts, associates, and founders doing their own work faster. Plans are credit-based and start at USD 18 a month.
Have us build a workflow
If your business runs the same process every week, we can build it to run on Smalt: the steps, the systems it reads from and writes to, and the points where a person has to sign off. We scope it with you on a call, quote it, build it, and hand it over with your team trained to run it.
Questions
Common questions.
What is Smalt AI?
Smalt AI is your AI coworker — one virtual employee that pulls up a desk and finishes real work. Research with citations, drafts emails, finds contacts, builds Excel models with real formulas, generates consulting-grade decks, and handles spreadsheet analysis. Tell it what you need in plain language; it does the work end-to-end.
How is this different from ChatGPT or Claude?
ChatGPT and Claude are general-purpose chat assistants — you talk, they answer. Smalt AI is purpose-built to finish knowledge-work deliverables: real Excel files with formulas, real PowerPoint decks, sourced research, contact enrichment, drafted outbound — handed back as artefacts you can ship. No prompting tricks, no learning curve, no agent orchestration on your side.
What can the AI actually do?
General-purpose knowledge work. It drafts emails, finds and enriches contacts, runs web research with sources, analyses spreadsheets, builds Excel models with real =SUM formulas (not LLM-typed numbers), generates Bain/McKinsey-quality PowerPoint decks, and connects to 850+ optional tools through Composio.
Which AI models power it?
Anthropic Claude (Sonnet, Opus, with 1M-token context for large documents) plus Google Gemini for fast classification. We pick the right model for each task — you do not manage that.
How does pricing work?
Credit-metered. The Free tier includes 500 credits and no credit card. Basic and Pro tiers add monthly credit allowances. Enterprise is self-hosted with custom pricing. Output tokens cost more than input — we follow the underlying model providers' pricing math, no markup gimmicks.
Do you build custom workflows for businesses?
Yes. If your business runs the same process every week and it touches a few systems, our team can build it to run on Smalt. That covers the automation itself, the integrations with your ERP, CRM, or document stores, a coworker set up for one team, and the pilot, training, and support to get it running. It starts with a call and a written scope, and is priced against that scope. Details at smaltai.com/services.
Is my data private?
Sub-processors are listed at smaltai.com/legal/sub-processors. Classifier calls run under Google's enterprise terms (no model training on your data). For teams that need full data residency, Enterprise self-hosting is available.
Pull up a desk.
500 free credits, no card. If you would rather talk first, write to us and we will come back to you directly.