The AI reads the request
A language model classifies what a trader asks and proposes a structured strategy record. It never executes, never sees a broker or a key, and never fills a gap silently. The details are on the product page.
About
We are building the part of trading software that is hardest to fake: understanding.
Connecting to a broker, signing in and taking payment are solved problems. Knowing that a system understood a trader's idea, and can show it, is not. That is the work we chose.
Mission
Make a trader's idea precise enough to test, and honest enough to trust.
We build the intelligence, buy the plumbing and design the boundaries. The language model reads what you mean. Deterministic tools calculate. A compiler formalises. Validators verify. You approve. A risk engine controls. Each step is separate on purpose, so that a mistake in one cannot hide in another.
The first customer is one trader describing one strategy, with no technical background. If they cannot tell what the system understood, we have not finished.
What we believe
Four rules we hold ourselves to.
Comprehension before automation
What matters is that the system understood you, and can show it. Automation without understanding is just a faster way to be wrong.
Nothing silent
Every value you did not state is recorded as an assumption and shown to you. The AI never fills a gap quietly, however helpful that would feel.
Proof before permission
Verification comes before approval. A check that could not run is reported as skipped, never as passed, and a strategy that took no trades is a failure, not a clean result.
Boundaries by design
The model cannot trade and the risk engine cannot be overruled. We enforce that with structure and tests, not with a policy document nobody reads.
How we work
AI is in the product, and in how we build it.
Both uses follow the same principle: a model proposes, something independent checks, and a person decides.
AI agents build and criticise
We use AI coding agents (currently Claude Code) as engineering partners. One agent builds a change, another tries to break it. Critics run the tests and mutation checks rather than just reading the diff, and a change is not done until the critics are satisfied.
A person decides
Our rule: when an AI coding agent works unattended it stays on local branches, and nothing is pushed or merged until a person has reviewed it. Paid evaluations against the live model are run in small, scoped steps, and we say what a run will cost before we start it.
Of the 433 commits in our repository on 11 October 2026 (merge commits not counted), our AI coding agent authored 325 and is named as co-author on 431. That is the reason review matters.
History
A few weeks, in order.
CTE Algo is young. This is the real sequence, from our own records.
The idea on paper
The first architecture documents: an AI-powered trading platform for people with no trading experience, built around conversation.
The repository starts
A strategy specification, a tool registry and the first deterministic tools. An execution core is proved against a broker demo account in the days after.
The rules are written down
The architecture is consolidated around invariants: the model never executes, has no broker or secret access, and never fills a gap silently. The first measured run of natural language to strategy follows the same day.
Checks before approval
A verification stage that refuses a strategy fitted to a random walk.
Draw it instead
Visual strategy mapping: draw on the chart and get a candidate strategy, found by deterministic geometry.
CTE Algo
The product takes its name and a new workspace, with Paper and Night themes, replay, signals and the assistant in one place. It is being tried behind a switch.
Team
A small team, on purpose.
We keep the group small so everyone can read the whole system. AI coding agents sit alongside as builder and critic.
AI coding agents
Builder and critic agents that work alongside the team. A person reviews their changes before they merge.
At a glance
The company in five lines.
- Name
- CTE Algo
- What we make
- An AI-native workspace for building, checking and testing trading strategies
- Stage
- Early access. Started September 2026
- Web
- ctealgo.com
- support@ctealgo.com
Want to work with us, or try it?
Request early access, or write to say hello. Partnerships and press are welcome too.