Midnight Presents Companions AI Agents for DeFi

The Midnight Foundation showcased Companions at its Midnight Cafe event, presenting an AI assistant designed to find services based on its owner’s goals and preferences. Gareth Christian Lim, the foundation’s director of growth, demonstrated purchase discovery, market comparisons, and post-trade analysis, while outlining how the system could connect users with applications across blockchain networks.

Companions Finds Services and Tracks Investments

One demonstration began with a request to buy $NIGHT. The agent checked market information and searched the ecosystem for a suitable purchase venue. Lim described how connections to Midnight Passport and SIG Network’s cross-chain infrastructure could extend that search to other networks.

The proposed workflow brings service selection into the conversation with an agent. Users specify what they want to accomplish, while Companions identifies relevant applications and providers.

A second example compared market information, including trading data from different sources. Lim described responses tailored to the user’s interests, whether those involve research, trading, or token swaps.

The third demonstration followed an investment after purchase. Using a token bought 90 days earlier, the agent presented information about its subsequent performance. Lim also described proactive updates, allowing Companions to surface relevant developments without waiting for its owner to ask.

A Familiar Coffee Shop Explains the Personalization

Lim illustrated the product’s approach through a coffee shop he regularly visits in Thailand. Its staff recognize him, remember his loyalty account, and know his preferred order, including how it changes later in the day.

He applied that relationship to Companions: an assistant should retain enough context to make future interactions more useful. The intended personalization includes a user’s goals, interests, and risk tolerance, helping the agent select information and services suited to that person.

According to Lim, those preferences are intended to remain in private state accessible to the user and their agent. Privacy is part of the proposed design for handling the personal context that informs its responses.

The same approach would help users discover relevant ecosystem projects. Companions could filter incoming information and identify providers matching its owner’s interests, a function the foundation connects with customer acquisition and retention.

The presentation did not specify a general availability date for Companions.

Passport Demonstrates Limits on Agent Purchases

In a separate presentation at the event, Midnight Passport demonstrated controls for authorizing financial actions. The demo used Grok to show how an agent could operate under permissions established by its owner.

Under the presented model, a user sets rules and grants limited authority. The agent receives an allowance rather than the user’s master key, while a counterparty can verify that an action falls within the authorized scope.

Passport was presented as a connection between identity, wallets, and applications. Lim placed it alongside Companions and SIG Network in Midnight’s broader tooling strategy; the Passport demonstration focused specifically on enforcing permissions.

After connecting Passport to Grok, the presenter instructed the agent to buy ETH. A purchase was shown, followed by a further request that was rejected under the configured restriction.

The presenter then revoked the agent’s access using a passkey. A subsequent instruction to buy ETH was denied because the agent no longer had authorization.