Desk2Quant Quant Agent CLI v2

A terminal-native quant copilot for learning, rigorous problem solving, interview practice, adaptive assessment and project design.

Requirements

• Node.js 20+
• A paid Desk2Quant purchase using the same email you sign in with

Install

npm install -g @desk2quant/cli

For development from this repository:

cd cli
npm link

Sign in

d2q login [email protected]

Desk2Quant emails a My Access sign-in message. For CLI authentication, copy the raw https://desk2quant.com/my-access.html?email=...&tk=... URL shown in the dedicated CLI box and paste it at Magic link >.

The CLI exchanges that temporary sign-in credential for a signed Quant Agent session. Model, Razorpay, Supabase and email-service secrets remain server-side.

Launch the dedicated TUI

After sign-in, run:

d2q

This opens the Quant Agent v2 terminal interface. Ask questions naturally; the agent automatically routes the request to the appropriate workflow.

Examples:

Explain why gamma becomes concentrated near the strike as expiry approaches.
Derive the Black-Scholes PDE from a delta-hedged portfolio.
Interview me for an equity derivatives quant role.
Design a production-grade Heston calibration project.
Give me a probability drill focused on conditional expectation.

TUI controls

/mode auto|learn|solve|practice|interview|project
/depth concise|standard|deep
/progress
/skills
/assess <skill>
/submit <assessment-id> <answer>
/context
/history
/sources
/new
/export
/clear
/help
/quit

The default mode is auto, and the default answer depth is deep.

Multi-turn context

The TUI keeps recent conversation context locally so follow-up questions can refer to the current discussion. For example:

You: I am calibrating Heston to an SPX smile.
You: Why is rho unstable?

The second question is interpreted in the context of the Heston calibration discussion rather than as an unrelated definition of rho.

Local TUI history is stored under ~/.desk2quant/ with owner-only permissions where supported. /new starts a fresh conversation and /export writes the current transcript to a Markdown file in the working directory.

One-shot commands remain supported

The v1-style commands still work for scripts and quick terminal usage:

d2q learn "Ito's lemma"
d2q solve "derive E[S_T^2] under GBM"
d2q practice "conditional probability"
d2q interview "quant research"
d2q project "Heston calibration"
d2q progress
d2q whoami

Adaptive assessment

d2q assess probability
d2q submit <assessment-id> "your answer"
d2q skills

theta is a bounded latent ability estimate updated from graded assessments. It is not a percentage or percentile.

Supported assessment areas include probability, linear algebra, statistics, stochastic calculus, derivatives, fixed income, numerical methods, programming, risk and quant research.

Private Desk2Quant retrieval

When the signed-in buyer has a Razorpay-verified entitlement to an indexed Desk2Quant product, relevant excerpts from that product can be retrieved server-side and used as grounding. The CLI cannot query the private corpus directly and no paid-book text is bundled with this package.

The TUI does not fabricate citations. /sources only displays source metadata when the backend actually returns it.

Local credentials

The CLI stores only the signed Desk2Quant agent session in ~/.desk2quant/config.json, created with owner-only permissions where supported.

To remove the local session:

d2q logout

Security

• No Desk2Quant model/API secret is shipped in the npm package.
• Purchase entitlement is verified server-side.
• Authentication uses short-lived My Access credentials and a signed local agent session.
• Conversation history is local to the CLI unless included in a user request sent to the Desk2Quant backend for contextual answering.