Your coding agent can trade now.

Run Claude Code, Codex, OpenCode, Pi, and more against one self-hosted trading desk. Agents research, schedule, and stage operations; you decide what reaches the broker.

01ask alice / pi session / webpi

Give your agent the brief.
Ask Alice launching a real Pi research Session, then opening the same Session in WebPi to review its three-step plan

The operating model

Give the agent a system,
not another chatbox.

One Codex Session turns a brief into a named Issue. The same object keeps its schedule and run history, then publishes a source-bound report to Inbox.

Durable handoffSame Codex Session

The brief becomes an object.

Codex leaves one named Issue with its owner, schedule, instructions, and creation provenance on disk.

Codex creatorStable Issue idSession ownership
Read the workflow
OpenAlice opening a scheduled NVDA research Issue and revealing the same Codex Session that created it

Specialist research

Go deeper.
Keep the same workflow.

Give the same agent focused environments for experiments, evidence, and the next research decision.

Quantitative researchGoverned research workflow

Give the agent a laboratory.

Run projects, studies, and experiments in one inspectable research workspace.

Project → Study → RunValidation gatesAuditable workspace
Read the workflow
OpenAlice opening the AutoQuant Studio to inspect a research project, validation metrics, and its next experiment

Reviewable execution

See the trade
before it moves.

Inspect the account, side, quantity, order type, and operation diff before anything reaches a broker.

Trading as GitNeeds approval · no broker write

The broker waits for you.

The agent prepares the operation. You inspect the exact diff, then a second human confirmation stands between the proposal and the broker.

Agent proposesOperation diffHuman push
Read the workflow
OpenAlice Trading as Git showing a staged BTC demo order, its operation diff, and the first approval click stopping at a second confirmation without contacting the broker

Built in the open.
Kept on your machine.

AGPL-3.0 · Local-first · Bring your own agent

Before you install

Questions,
answered plainly.

Open source does not have to mean guesswork. Here is what runs, where it runs, and who stays in control.

What is OpenAlice?

OpenAlice is a self-hosted trading workspace for coding agents. It gives an agent a place to research, keep files and memory, run scheduled work, and optionally prepare broker actions through one reviewable workflow.

See the product overview
Which coding agents can I use?

Bring the agent CLI and subscription you already use. OpenAlice supports Claude Code, Codex, OpenCode, Pi, and other compatible agent CLIs without forcing you into a bundled model plan.

How agent sessions work
Will an agent place trades without asking me?

Not by default. Alice can stage and commit an intent, but broker execution waits for your approval in the Web UI. Direct AI push is an explicit Pro setting; start with the simulator, paper, demo, or testnet path first.

Read about the approval wall
Do I need a broker account to get started?

No. Research, watchlists, issues, workspace automation, Inbox, and the built-in mock simulator work without trading credentials. Connect a broker only when you are ready to validate execution.

Start with the quick start
Which brokers and markets are supported?

Optional Broker Packs cover paths including Alpaca, CCXT exchanges, IBKR, Longbridge, and LeverUp. Venue, instrument, and order-type support varies, so validate your specific route with paper or demo funds before going live.

Check broker details
Where do my data and credentials live?

OpenAlice runs on your machine and keeps its product state locally. Broker credentials are sealed at rest with a machine-bound key; the setup wizard is the normal way to add or change them.

Review data and credentials

Open source · self-hosted

Get OpenAlice.

Download the local desk, then bring the coding agent and subscriptions you already use.