AI designed for compliance, speed, and control
Knapsack builds AI agents for financial services firms, including RIAs, wealth managers, private equity and VC, and asset managers. Powered by Claude, its agents automate client-ready reports and daily work across email, calendar, documents, CRM, and portfolio systems, from meeting prep and notes to CRM updates, email, and document research.


Financial services teams spend hours moving information across disconnected CRMs, portfolio systems, email, and documents, while navigating strict compliance and data residency requirements.
Knapsack started with private, accurate meeting transcription for financial advisors. Today, its multi-tenant SaaS platform uses Claude-powered agents to work with each firm's data inside its own cloud environment.
From local AI limits to hybrid, compliant intelligence
Knapsack's early desktop app ran AI locally to protect sensitive data, but this limited performance and made advanced capabilities and integrations difficult to scale.
Knapsack evolved to a hybrid, privacy-first architecture that maintains data control while enabling the speed, automation, and integrations modern financial firms need.

Re-architecting for speed, trust, and broader functionality
Since May 2024, Cheesecake Labs has worked as Knapsack's core engineering team: a dedicated, cross-functional squad of backend, frontend, AI, and DevOps engineers working alongside the founders.
The squad owns features end to end, from architecture to production deploys, across the agent engine, the connector layer, the Claude infrastructure, and the web app.

The partnership aimed to turn Knapsack into a general AI agent platform for financial services — one where users chat with an assistant or configure reusable, schedulable Workflows that act on their real data, without compromising on compliance.
To get there, Knapsack needed a model that could reliably reason over many steps, use tools across dozens of third-party systems, and write output accurate and professional enough for regulated firms. In 2025, Claude on AWS Bedrock became the platform's core engine.

Automated client reports: from scattered data to client-ready documents
Producing a client report usually means pulling numbers from a portfolio platform, context from client documents, and hours of writing and formatting. Cheesecake Labs built a Claude-powered report Workflow that does this automatically. It:
Combines portfolio data from platforms like Tamarac with client documents stored in SharePoint
Writes client-ready reports in the professional tone high-net-worth clients expect
Records which document or data source each section came from, so every report can be audited
Keeps report versions per user and runs on demand or on a schedule
Claude plans tasks, selects tools, runs multi-step actions, requests approval when needed, and delivers sourced results.
Firms automate recurring tasks such as CRM updates, meeting follow-ups, LP reports, and email.
Agents are available in the Studio, Knapsack Chat, an embeddable AI widget, and a Chrome extension.
20+ custom MCP connectors connect Claude to Salesforce, Wealthbox, Tamarac, Microsoft 365, Google Workspace, and more.

Baking compliance into every layer
Compliance and privacy remain at the heart of Knapsack's value proposition:
- SOC 2 Type 2 compliance
- HIPAA compliance for sensitive data handling
- GDPR-aligned practices to meet global privacy requirements
- Connected customer data is never used to train models

For enterprise customers contractually restricted to AWS, we built a fail-closed, Bedrock-only mode: every Claude call, including background jobs, stays inside AWS Bedrock, and any call to another provider raises an error instead of falling back. Compliance journaling records every LP-facing chat interaction.
Knapsack also offers self-hosted environments, giving enterprises complete control of sensitive data and resilience against third-party service risks.

Core platform improvements at a glance
To balance performance, cost, and privacy while expanding functionality, Cheesecake Labs implemented:
01
Automated client reporting
Claude-powered Workflows that turn portfolio data and client documents into auditable, client-ready reports.
02
Agent orchestration engine
Claude plans each request, selects tools, runs multi-step tool calls, and pauses for approval when needed.
03
20+ MCP connectors
An MCP orchestration layer for CRMs, portfolio platforms, email, calendar, and document stores.
04
Role-based model routing
Claude Haiku 4.5 for fast orchestration steps, Claude Sonnet for final answers, and prompt caching to keep costs low.
05
Resilient Claude infrastructure
Retries, an ordered fallback chain up to Claude Opus 4.7, and per-operation token and cost tracking.
06
Enterprise compliance controls
Bedrock-only enforcement per organization, compliance journaling, and a private RAG pipeline for documents.
Proven results: what customers are achieving with Knapsack
AI agents in daily operations at regulated firms Knapsack now runs in production with paying customers in venture capital, wealth management, asset management, and fintech — firms that rarely put AI tools into their daily operations. Its Claude-powered agents have processed more than 3 billion tokens of CRM records, portfolio data, emails, and documents.
Performance and cost under control at scale Routing simple, high-volume steps to Claude Haiku and final answers to Claude Sonnet, with long system prompts cached, keeps multi-step agents fast and affordable — even when a single request takes up to ten plan-act iterations.


Productivity, compliance, and scale — delivered
The same Claude-based engine now powers very different use cases, with only configuration and connectors changing per client:
- Wealth management: automated, auditable client reports built from portfolio data and client documents, with every section traced to its source.
- Venture capital: a Claude-powered assistant that a deal-flow platform offers its own clients, under its control and fully auditable.
- Asset management: a compliant LP chat that runs entirely inside AWS Bedrock, with every interaction journaled.
Knapsack has adopted each new Claude generation soon after release — from Claude 3 to Claude Sonnet 4.6 and Opus 4.7 — without a platform rewrite. Our team also builds the product with Claude Code, which has co-authored more than 200 commits in the Knapsack codebase.

Here's what Knapsack has to say about working with us.
Knapsack's leadership immediately noticed how much simpler and more effective their operations became. Here's what they think of our partnership:
"We couldn't have solved some of the complex problems to deliver this product without the support of Cheesecake Labs' team."

Mark Heynen
Co-Founder & CEO @ Knapsack


