plan.ai labs · engineered for trust

An autonomous trader built on AI councils.

Language models propose. Deterministic, unit-tested code disposes. Every decision it makes, and every one it refuses, is logged and replayable. It earns autonomy by proving itself first. Paper-only for now, and it claims no returns.

Safety is code, not prompt Earned autonomy Multi-model council Paper-first
decision receipt#a7f3…e2b9
proposalBUY NVDA · 120 · conf 0.71
universeapproved & ethical
point-in-timeno look-ahead
long-only / typepermitted
risk enginegross-exposure cap exceeded
⊘ BLOCKEDno order placed
reason · would breach 2.0× portfolio cap (policy < broker max)
logged · 2026-06-01T14:32:07Z · replayable from snapshot

A real decision artifact, including the ones it refuses.

Why trust starts here

Most systems chase returns. This one earns trust first.

Returns are easy to claim and hard to prove. The architecture is the part you can actually verify, so we built these five things into it from the start.

01

Safety is code, not prompt

Language models propose. Deterministic, unit-tested code disposes. Every capital decision passes hard gates, and no model reaches the broker directly.

02

Decisions survive a council

Claude, OpenAI codex, Gemini and GPT cross-examine each call before it stands. Adversarial by design, a hedge against any single model's overconfidence.

03

Replayable to the snapshot

Every decision, and every refusal, is hashed and trace-linked. You can rebuild it from the exact point-in-time data the agent saw.

04

Designed fail-safe under leverage

Margin, leveraged ETFs and defined-risk options each sit behind deterministic caps. Auto-deleverage and broker-side resting stops are built to hold through a full outage.

05

Honest by construction

It runs on point-in-time data with no look-ahead and grades itself on drawdown. The point is to prove the machine works before claiming any edge, and to say which is which.

Earned autonomy

It earns its independence one proven guardrail at a time.

The goal is an agent you trust to trade on its own. It gets there on evidence: each level unlocks only after the audit trail proves the one before it. Risk-reduction runs automatically from day one, and the deterministic rails hold at every level.

now
L0
Human-gated
Every entry needs your sign-off.
L1
Auto risk-off
Exits & de-leverage act on their own.
L2
Auto low-risk entries
Pre-cleared setups, inside caps.
L3
Supervised autonomy
Acts within proven envelope; you watch.
Proof of process

It is as proud of the trades it refuses.

Anyone can show a winner. The rarer thing is a system whose limits live in code, not in a pitch. These are the rails it runs on.

It won't let a language model place an order. 01
It won't trade beyond the autonomy you've granted it. 02
It won't run leverage without a broker-side stop. 03
It won't use a data point it could not have known in time. 04
It won't hide a decision, or a block, from the audit log. 05
It won't claim a return it has not earned. 06
The central design line

Models propose. Code disposes.

One boundary, place_order, separates judgment from authority. No model output reaches the broker without passing deterministic, tested gates.

left of place_order

Models think

  • persona screening
  • bull / bear / risk debate
  • sentiment, theses, proposals
place_order
right of place_order

Code decides

  • risk & leverage engine, sizing
  • ethics, kill-switch, double-gate
  • broker-side stops, fail-closed audit
No single model decides

Every decision survives a council.

No lone genius model. Independent models cross-examine each call, and the disagreement is logged before the decision stands.

Architect
Claude
Skeptic
codex
Critic
Gemini 3.1 Pro
Pragmatist
GPT-5.5
No spin

What this is, and what it isn't.

Status: paper-first. The v1 goal is to prove the machine, not the edge: a fully-audited paper-trading loop on real, point-in-time market data. No returns are claimed or implied. This lab intends to earn and measure performance, not advertise it. Internal proprietary research, not investment advice. Austria / EU (WAG 2018, FMA, MiFID II, MAR).
Verify it yourself

Don't trust the pitch. Read the architecture.

Every claim here is documented and council-vetted. Go read it.