The questions advisors ask first.

Those tools model wealth trajectories. Foundry models cash flow year by year on a deterministic ledger. The numbers you actually discuss with a client (Roth conversion dollars, bracket fills, withdrawal sequencing, RMD timing) are primary inputs, not approximations layered on at the end. Federal brackets, capital gains stacking, AMT, NIIT, Social Security taxability, IRMAA, and state tax for every state and DC all resolve on that same ledger, and the same engine drives the Monte Carlo. Same inputs, same numbers, every run.

Each projection year, Foundry computes federal graduated bracket income tax, long and short term capital gains with 0/15/20% preferential stacking, AMT with exemption phase out, the 3.8% NIIT, the QBI deduction, Social Security taxability, FICA and self employment tax, IRMAA Medicare premiums on the real two year MAGI lookback, and state income tax for all 50 states and DC. It also models RMDs, Roth conversions, withdrawal sequencing, and trust 1041 tax. Taxes post to the year's ledger as the Taxes line, traceable down to the account that generated them.

Yes, all on the same year by year ledger, so they interact the way they actually do. Roth conversions support bracket fill targeting that accounts for Social Security taxability and how QBI phases in; RMDs use SECURE 2.0 start ages and the Uniform Lifetime Table; IRMAA applies the real two year MAGI lookback, so the cost of a conversion or a large RMD surfaces two years downstream; and AMT and the 3.8% NIIT are computed in the same pass. Withdrawal sequencing follows your priority order with the correct tax character drawn from each account.

Every Monte Carlo trial reruns the full Foundry projection engine, the same one behind the deterministic plan, overriding only each account's annual growth with a correlated market return drawn from your firm's capital market assumptions. Income and capital gains tax, RMDs, Roth conversions, Social Security, IRMAA, and withdrawal sequencing all run unchanged on every trial, so the probability of success reflects real tax drag, not a returns only wealth line. The default is 1,000 paths, and each scenario stores its own seed, so the same plan returns the same number every time.

Yes. Foundry computes federal estate tax on the Form 706 lines, including spousal portability of unused exclusion (DSUE), against the current $15M basic exclusion amount, which the engine treats as permanent under OBBBA, indexed for inflation, with no 2026 sunset. State estate tax is modeled for 13 jurisdictions and state inheritance tax for five: Pennsylvania, New Jersey, Kentucky, Nebraska, and Maryland. A per spouse gift ledger tracks annual exclusion and lifetime exemption use and warns before a gift breaches the exemption. It models ILIT, IDGT, CLT, and CRT structures, installment sale notes, life insurance routing with the three year lookback, and shows each beneficiary's projected inheritance.

Yes. A dedicated Stock Options account type covers RSUs, NQSOs, and ISOs with grant tranches, 83(b) elections, and a forward exercise and sell plan: RSU vests and NQSO spreads are taxed as ordinary income, ISO exercise feeds the AMT engine as a real preference item, and sales realize gains by holding period. ESPPs are not yet modeled. A business is modeled as its own asset that grows and can own child accounts, with a distribution policy and pass through taxation to the owner's 1040; C corporations and foundations are currently simplified to pass through treatment.

Not without you. Forge is an agent inside the app: 67 tools that read client data, run the engine, draft scenarios, and assemble the deck. Every plan data and scenario write (accounts, income, expenses, liabilities, scenarios) surfaces as an approval card you confirm or reject before anything saves, and conversations are checkpointed so any turn can be undone. Reversible practice management writes, such as adding a note or a task, apply immediately and are audited. Answers are grounded: when a reply states a financial figure, a deterministic check traces it back to an engine result and a second model audits the numbers before you see them. Where that check is not satisfied, Forge flags the answer as unverified rather than presenting it as settled. It is a drafting tool that can still be wrong, so review its output before it reaches a client.

It depends on the path. If you upload documents, the extractor sends them to Azure OpenAI, which does not train on them, and returns structured data to a review queue you confirm before anything touches a plan. Social Security numbers are stripped from a document's text before it is sent; scanned pages and photos are sent as images, so a number printed on them goes too. If you connect Orion, Foundry makes a read only OAuth pull of accounts and holdings and never writes back. If a client uses the portal, they can link accounts through Plaid, and those balances update the plan without a review step. If you enter data by hand in the onboarding wizard, nothing leaves at all.

No. Neither our document extraction nor the Forge agent trains on your data. Both run on Azure OpenAI, which does not use prompts or outputs to train models, and your plan data is stored in US regions. Before document text is sent for extraction, Social Security numbers are pattern matched and stripped; scanned pages are sent as images. Forge is instructed to keep its long term memory to nonsensitive preferences and recurring assumptions, never dollar figures, account numbers, or financial detail, and none of it becomes training material.

Every client data query is scoped by your firm's ID on the server, derived from your Clerk session rather than from anything the browser sends. An automated isolation test scans the client, CMA, CRM, and portal API routes for that scoping and confirms none of them is cached. A request for a client outside your firm returns the same result whether or not the client exists, so existence is never leaked. Inside the app, Foundry staff reach firm data only through support sessions that require a written reason and are logged, opened by a sign in link that expires after ten minutes.

Not yet. Foundry does not hold its own SOC 2 attestation, but it runs entirely on infrastructure that does. Neon, Vercel, Stripe, Clerk, Sentry, Upstash, and Azure OpenAI are all SOC 2 Type II attested, and most carry ISO 27001 as well. We have completed an internal SOC 2 Type 2 readiness assessment that mapped the controls and the remaining work, and pursuing our own attestation is on the roadmap rather than finished. Today we can share our security overview, subprocessor list, and DPA; we give 30 days' notice before adding or replacing any subprocessor that handles client data; and application data is stored in US regions.

No separate CRM required. Foundry has a native, household based CRM in the same app as the planning engine: contacts, accounts with custodian, type, and balance, an activity log for calls, meetings, and emails, notes, tasks with due dates and assignees, and a versioned document vault. A firm is one organization; you add advisors as members and bill per seat. Every firm member sees the whole book today, and you can share an individual client or your entire book, at view or edit level, with another Foundry advisor, who can be at a different firm. Forge can also handle CRM work like meeting prep, note summaries, and follow up drafts.

A clean, branded PDF deck assembled from a library of 42 single focus report pages: cover, cash flow and its drill downs, a full income tax breakdown, balance sheet, asset allocation, estate transfer and gift tax, Monte Carlo, life insurance, Medicare and IRMAA, and scenario comparisons. You choose the pages, set each one's options, reorder them, and generate. The deck renders server side in your firm's logo and accent color and lands in the client's document vault. The client gets the decision, not a hundred page binder.

Onboarding is about getting real client data in, then building the first plans on it with you. Data comes in three ways: a direct Orion household sync, Plaid account aggregation through the client portal, or a guided upload where our importer extracts accounts and holdings from PDF statements, Excel, or CSV. Our team builds those first plans alongside you so you watch the engine run on your own numbers. From there it is built for live work: multidecade deterministic projections recalculate as you change inputs, a 1,000 trial Monte Carlo is one click from the solver, and a finished plan exports to a branded, firm logo PDF.

One flat plan, billed per advisor seat: $99 per advisor per month, or $999 per advisor per year ($83.25 a month, a 16% saving) when billed annually. Every feature is included, with no tiers and unlimited clients per advisor, and AI document import carries no add on or per import charge. The public 14 day free trial requires a card and converts to a paid subscription unless you cancel first, and you can cancel anytime from the billing portal in Settings. Prices are in USD and exclude tax.