Skip to content

Flagship product

Quant Agent

A planned read-only quant research agent for quantitative specification derivation and math/code audit. The assistant is not yet connected; no model has been called.

Quant Agent

Derive the spec. Audit the reasoning.

A planned read-only quant research agent for quantitative specification derivation and math/code audit. The assistant is not yet connected; no model has been called.

Read-only · deny-by-default

Research workspace

PLANNED · NOT YET CONNECTED

Assistant not yet connected. No model has been called. No repository access is connected.

Checking account…

How it works

Planned commands and read-only scope. None are executable in this interface today.

CommandJobModelApprox. / invocation
q planPlanned: derive new math — formula, units, a point-in-time proof, a written spec, and an explicit falsification criterion.Not yet assignedNot yet provided
q auditPlanned: verify math or code handed to it, using read-only commands.Not yet assignedNot yet provided
q second-opinionPlanned: challenge a conclusion the agent has no stake in.Not yet assignedNot yet provided
q pingPlanned: verify connectivity with one real call once a model backend is connected. No call is made today.Not yet assignedNot yet provided

Product invocation costs are not yet provided. Internal precursor costs are not quotes for the planned lineup. No model is connected or called today. Private beta is free; nothing is billed today.

Read-only is a design property.

Read-only and deny-by-default is the intended product design. The assistant has no model backend or repository access today. It cannot write to a repository, run training jobs, access production databases, execute arbitrary commands, or place trades.

The internal precursor tested the absence of arbitrary command execution. This is not evidence of a connected product runtime; the read/inspect tools below describe intended scope only.

Planned read/inspect tool surface · not connected

  • read_file
  • list_dir
  • glob_files
  • grep
  • git_status
  • git_log
  • git_diff
  • git_show

Planned model lineup

PLANNED · NOT YET CONNECTED
Claude (Anthropic)proprietary-hosted
Intended hosted research model; command assignments are not yet defined.PLANNED · NOT YET CONNECTED
Open-weight families: Llama, Mistral, Qwenopen-weight
Intended to run locally or on your own infrastructure. Families indicate direction, not supported checkpoints.PLANNED · NOT YET CONNECTED

Propose, challenge, verify.

  1. plan

    Proposes

  2. Primary agent

    Critiques and implements

  3. second-opinion

    Red-teams

  4. audit

    Verifies

Planned workflow, adapted from the internal precursor. Implementation belongs to a separate primary agent, not this read-only product. A shared model can propagate a misreading between plan and audit; an independent second opinion is intended to challenge that conclusion. Product model assignments are not yet defined.