What is oppen?
oppen is a local-first trading terminal and execution gateway for AI agents. You choose the agent, allocate capital and define its permissions. The agent requests actions; the Rust core is designed to check policy before signing them. The console makes execution and operator intervention visible. The project is pre-alpha with no public release.
Is oppen AI infrastructure or a trading bot?
It is infrastructure for agent trading. The terminal is its first product: a local place to manage permissions, signing, account state and supervision. Oppen does not provide a model, a default strategy or a claim of profitable signals. Agents use it to act; operators use it to retain control.
Does oppen operate an exchange?
No. Orders are sent to an external venue, which holds the margin account and determines execution, funding and liquidation. Oppen provides the client and execution controls. Hyperliquid is the first target; using Oppen does not remove the risks or requirements of using that venue.
Can I use it with real money today?
There is no public release and no claim of live-trading readiness. Read and execution code exists in development, but venue acceptance, recovery and the full operator workflow still have open gates. Testnet is the default. The status page distinguishes implementation progress from demonstrated readiness.
Do guardrails guarantee a maximum loss?
No. They are designed to constrain requests before signing and stop further activity when configured conditions are reached. A loss circuit breaker reacts to measured loss; it cannot guarantee an exit at that value. Market gaps, fees, funding, liquidation and connection failures can produce losses beyond a trigger. Halting new orders does not close existing positions.
Who holds the keys and funds?
Funds and margin remain in your venue accounts. The intended key model keeps account-owner private keys in your own wallet. Oppen generates separate agent-wallet keys locally and uses them in the Rust signer, backed by the OS keychain. The Oppen team does not receive those keys. This is different from saying the local application never handles a key: signing necessarily uses one.
What if an agent is prompt-injected or compromised?
Requests made through Oppen remain subject to the agent's assigned policy, assuming the core and its state are functioning correctly. Harmful trades can still fit that policy. If an attacker obtains the signing key or compromises the host, it can bypass local controls and trade directly. Only the venue's own key permissions remain; a compromised account-owner wallet can expose still more authority.
Can I connect my own model or scripts?
That is the intended integration model. Clients connect through local MCP, and developers must follow the tools and configuration exposed by their build. A hosted assistant is not automatically able to reach a local endpoint. The model or script remains responsible for its strategy, and its client controls what context is sent to a model provider. See For agents.
How are different agents separated?
The design assigns each agent a dedicated venue account, called a container. It may be a sub-account when available or a separate top-level account. v1 plans for dedicated top-level Hyperliquid accounts. Each account requires its own capital allocation and applicable approvals. Separation at the venue does not isolate keys from an attacker who controls their shared host, and it does not combine margin across accounts.
What data does oppen collect?
The desktop application is designed to keep its execution records and configuration locally, without usage analytics sent to the Oppen team. It still connects to the venue; your chosen agent client may send context to a model provider. The website uses Cloudflare Web Analytics for visits and performance. If you email a launch-notification request, we receive that email. See Privacy for the separate data practices.
Which venues are supported?
Hyperliquid is the v1 execution target. Aster and Lighter are planned, not supported releases. Multi-venue routing and combined portfolio views are future work. Venue-specific permissions must be verified independently; no claim about a Hyperliquid API wallet should be assumed to hold on another venue.
What does it cost and how is it licensed?
The planned application model has no software subscription and uses a builder fee on eligible routed fills. On Hyperliquid the account owner approves a maximum builder fee. Venue fees, funding and model-provider charges are separate. The effective rate and approval should be reviewed before use; this pre-alpha site is not a final release price schedule.
The project is open core: the enforcement core is designated Apache-2.0, while the console uses a commercial source licence with different redistribution rights. Source publication is planned for launch. That does not make the whole application unrestricted open source or independently audited. See Terms.
Why the name oppen?
The name refers to Oppenheimer and the responsibility that accompanies powerful tools. The project's purpose is practical: make software acting with capital easier to constrain, inspect and supervise. That responsibility includes stating what is unfinished and revising claims when the evidence changes.