How AngelList Cuts Investment Paperwork From Hours to Minutes With Agents

AngelList builds financial infrastructure for venture funds, investors, and startups. Today, it has $200B+ in assets and 30,000+ startups funded on its platform, with $90B in money movement running through its infrastructure.

When a fund manager wants to invest in a startup, AngelList coordinates the paperwork over email: collecting contracts, confirming investment details, and handling routine follow-ups. AngelList built a multi-agent system with Mastra to shorten this work from hours to minutes.

Turning an email-first process into an agent workflow

Each investment triggers a cascade of email coordination: wire detail verification, SAFE document collection, hundreds of signature requests, and a constant back-and-forth across multiple parties. The work is labor-intensive even though the tasks are often straightforward.

Coordinating an investment over email can be manageable for a small fund but AngelList has had to repeat that work across many funds, with different operations teams coordinating with fund managers and startups. The steps could happen in different orders, so the system needed to respond accordingly when emails and documents were coming.

AngelList's internal workflow management system already moved work through defined stages and could call a model to summarize a document or extract terms. It could not handle the open-ended part: search for context, use several tools, assess what had happened in the workflow, and decide what step came next.

The POC and why Mastra was the right framework

The team had built POCs with other frameworks, and tried other platforms for parts of the process, but they eventually chose Mastra because it was TypeScript-based. It was also important that Mastra’s agents and workflows could call each other, and memory controls and evals came with the framework. AngelList could also self-host it.

"Mastra helped us build a solution quickly while keeping complete control of the underlying infrastructure and code. Memory, tool calling, workflows and observability came out of the box, so we could focus on our actual business logic instead of building agent plumbing. Plus, everything is in TypeScript, which our team already knows well." — Nick Gervasi, Senior Software Engineer

As an initial proof-of-concept, an engineer built an internal email-search agent using Mastra, Assistant UI, and Front's search API. Instead of vectorizing the company's email, he gave the agent the same search interface a person would use. Four out of five searches produced good results.

But retrieving an entire Front conversation could overflow the model's context. So he added preprocessing inside the tool to strip forwarding cruft and summarize each email. It slowed the search down, but made the largest conversations usable.

The POC was the first step towards a larger system. AngelList’s workflow management system remained the system of record and the engine moving work through operational stages. Mastra gave those stages an agent they could assign work to.

Four agents around one system of record

Events in Front trigger Mastra workflows, which hand work to four agents:

• A planning agent that kicks off each deployment workflow.
• A signature request agent that monitors Front for incoming signature requests and uses Browserbase to open the linked page and download the document. It then sends an alert to the person responsible for the next step.
• A linking agent that matches incoming signature requests to the correct deployment task by analyzing email content and document terms.
• An execution agent that follows AngelList's standard operating procedure for deployments, determining next steps and drafting routine response emails for the team to review.

AngelList’s internal workflow management system holds the terms, progress, and status for each deployment. Every agent action is logged as a comment on the task. This gives context to both: agents and people. The next agent can start from where the previous one stopped. And the customer team continues working in Front while the agents move context between systems behind the scenes.

Mastra handles the orchestration layer. The team wraps each agent in a Mastra workflow, even for simple single-step calls, because it gives them structured inputs and outputs plus a clean history of all workflow runs.

The broader stack includes:

• System of record: AngelList's internal workflow management system
• Email and event source: Front
• Agent orchestration: Mastra agents wrapped in Mastra workflows
• Models: Claude and Gemini, depending on the task
• Evaluation and tracing: Braintrust, with the team building tool-call mocks and a more consistent eval suite
• Infrastructure: self-hosted inside AngelList's environment

Early results and what's next

The system is now running in production, reducing the processing investment paperwork from hours to minutes.

AngelList team catalogs failure modes and addresses them systematically to further improve the agent’s accuracy. The goal is to bring the deployment success rate to at least 80%.