Knapsack

Agents do the work in Knapsack. It reads a user's email, calendar, CRM and documents and creates five starting agents. Inbox Autopilot drafts replies in the user's voice, Meeting Prep sends a brief 30 minutes before each call, and the notetaker transcribes and summarizes. Follow Up drafts emails per attendee, and Daily Coach reviews the previous day.

The financial services digital workers apply the same engine to defined jobs, such as building evidence packets and exception reports for an RIA compliance review, or a case timeline and narrative draft for an AML investigation. Users can also build their own agents in plain language. There is no workflow system or data asset underneath that would still be worth buying without the models.

Every agent works under the same approval rule: nothing sends without the user, and every draft and every action is approved first. Connections to email and calendar are read only by default, and reminders and reports are saved as drafts for review. Each digital worker routes its findings to a designated owner before any action and states what it will not do.

The advertising reviewer never approves or publishes material, the AML analyst does not clear alerts or file regulatory reports, and the planning analyst does not contact clients without adviser approval. The onboarding analyst leaves account opening, approvals, signatures and client communication with authorized staff, and each worker produces a traceable review packet. In the Chrome extension the agent shows its plan and can be stopped or changed at any point. Users can schedule workflows, such as uploading notes to the CRM after meetings, and what runs unattended is then theirs to configure.

Transcription, drafted replies, briefs and the digital workers' findings carry no published accuracy figure, test or validation method. The workers keep links back to the underlying records and produce review packets, so a reviewer can check each finding against its source. The research analyst separates source facts from interpretation, and the portfolio reporting analyst leaves official performance figures to the firm's validated systems. The terms disclaim any warranty that information in the service is complete, accurate or timely.

Two named chief executives are quoted. Daniel Lopez, CEO of Sevanta, explains why it embedded Knapsack Chat rather than exposing its data to outside AI tools, and Guillaume Decalf, CEO of Oui Financial, is quoted on the site. ModelFA delivers its outreach, coaching and notetaker assistants to advisers on Knapsack, and Schwab Advisor Services lists Knapsack Pro and Knapsack Enterprise in its provider marketplace.

Three advisers named Knapsack as a notetaker they were considering in the T3/Inside Information 2026 Software Survey. Knapsack's adviser case study headlines eight hours a week saved, with no named customer or method behind the figure.

Knapsack says emails, meetings and documents are never used to train AI models, by Knapsack or anyone else. When cloud inference is needed, only the minimum context goes to SOC 2 Type II inference providers, which delete prompts and outputs as soon as a response returns. Hosted content is kept 30 days by default, and Knapsack describes a zero knowledge design in which staff cannot read files or notes.

Desktop keeps everything on the user's machine, and the notetaker captures meetings without recording video. Email history, meeting notes, relationship data and agent configurations belong to the user rather than a model provider. None of this is written into the terms of use, which do not address training.

Under the privacy policy, hosted content is kept for 30 days by default unless the user deletes it sooner or pins it, and extended retention applies only when the law requires. Data is shared only with service providers under contractual privacy obligations, for legal compliance or safety, or in a reorganization with equivalent safeguards. Transfers follow the EU US Data Privacy Framework principles, and users exercise choices through a privacy address. Studio reads only the data a user authorizes, and Desktop keeps it on the user's machine. No reference to GLBA or Regulation S-P and no breach notice deadline is published.

A SOC 2 Type II attestation and HIPAA compliance are stated on Knapsack's home, compliance and pricing pages, and Knapsack describes its controls as independently audited. It cites regular penetration testing, encryption at rest and in transit, annual security training and a security program led by IT and security leadership. Sign in runs through Google or Microsoft accounts, and the compliance page gives IT teams a short explainer for when that sign in is blocked. No report, auditor, report period or trust portal is offered, and the compliance page lists further standards Knapsack is working toward.

As a software supplier, Knapsack holds no license and names no regulator overseeing it. Its compliance pages describe support for firms' own duties. Notes, activity logs and communications are archived in a tamper proof format for FINRA audits, non compliant content can be flagged, and the platform can be configured for SEC, FINRA and HIPAA requirements. The terms make each user solely responsible for any regulation that applies to its use. The exam readiness worker keeps a document index and gap register ready before a regulator asks, while counsel and compliance keep scope, interpretation and every external response.

Supervision is the design principle Knapsack describes. Each digital worker routes its output to a designated human owner and does not give advice or contact clients. Each also lists the decisions it leaves to people, such as approving vendors, determining suspicious activity or setting credit terms. Daily Coach rates a user's work and the relationship tracker flags contacts going quiet, and how either decides what matters is not described. No AI governance framework, named owner or fairness testing is published.

Under the terms of use, the user carries the risk. The service is provided as is, with no warranty that it is accurate, complete, secure or error free, and a limitation of liability follows. Users indemnify Knapsack and its affiliates, the user carries any regulatory compliance its use requires, and California law governs. Knapsack can suspend or end an account at its discretion without liability. The approval step before any send and the review packets record what a firm approved. No warranty or remedy applies when an approved output proves wrong.

Groq and AWS Bedrock are given only as examples of inference providers, and no model, model family or transcription engine is named. Knapsack presents model choice as something it keeps fluid, letting users move between models while their context stays the same. The models behind any agent at a given time are therefore not stated. No subprocessor list or change notice is published, and the privacy policy refers to service providers without naming them.

Agents connect to Gmail and Microsoft email and calendars, Salesforce, HubSpot, Wealthbox, Zoho, Copper, Google Drive, Dropbox, Slack, Notion and custom MCP servers. The CRM agent matches contacts, creates records where none exist, logs meeting activity and sets follow up tracking. Sevanta's MyDealflow feeds portfolio and pipeline data to an LP update agent.

The digital workers draw on whatever a firm authorizes, such as its case system for AML work, its deal room files and financial models for credit diligence, or its retention schedule and repositories for books and records. Knapsack can also run headless inside another company's product, as it does for ModelFA and Sevanta. No API reference, field mapping or archiving connector such as Smarsh or Global Relay is named.

Knapsack lets a firm choose where it runs. Studio runs in the browser with agents in the cloud and end to end encryption. Desktop runs the same agents on the user's own machine, where data stays on the hardware and nothing reaches the cloud unless the user chooses, and agents and settings move between the two. The research analyst worker is offered as a free desktop download for analyzing filings privately. A Chrome extension adds a side panel, and a headless mode embeds the agents inside another company's application. The cloud hosting region is not named.

Starter costs $50 a month with 100 credits, all five agents, unlimited email and calendar connections and both Studio and Desktop. The pricing page says that typically covers 20 to 30 hours of meeting transcription. Each financial services digital worker costs $500 a month after a 14 day free trial, with the first task free, and runs on demand or on a schedule. Team and enterprise plans, with custom credits, integrations, priority support and an account manager, are quoted. Users can start free without a card, and fair use limits apply during the trial.

Financial advisers and RIAs are the lead regulated market, with a case study, a Schwab Advisor Services marketplace listing and a ModelFA partnership aimed at them. The digital workers extend this to RIA compliance and operations teams, exam preparation, marketing review, AML and sanctions analysts in financial crime operations, vendor oversight, portfolio reporting, risk monitoring, investment research and private credit due diligence.

Knapsack also says plainly that it is not exclusively for advisers. The same platform is sold to founders, operations teams and other professionals, and Sevanta embeds it for venture investors. Accounts require a work email address.