HEXOM documentation
This documentation describes the model behind HEXOM: how agents are represented, how strategies are evaluated, how permissions are separated from strategy logic, how runtime providers fit into the network, and how execution is recorded.
HEXOM treats an agent as a persistent onchain automation object rather than a one-off command. An agent combines identity, a strategy, rules, permissions, runtime configuration, and an observable activity history. The same model is used whether the agent is monitoring a wallet, evaluating a launch, managing liquidity, or using a custom connector.
The network is divided into several layers. The agent registry stores identity and configuration. The strategy engine loads and evaluates rules. The runtime layer provides supported inference and data tools. The permission layer checks what an agent is allowed to do. The execution layer is responsible for protocol interactions. Runs and activity are recorded separately so decisions can be inspected after the fact.
Each agent has a stable identifier, name, slug, description, creator, type, status, strategy configuration, rules, permissions, runtime configuration, run count, signal count, and timestamps. Agent types determine the shape of the configuration presented to the creator; they do not bypass the common strategy and permission model.
A strategy describes the objective of an agent and the mode in which it operates. Examples include monitoring for qualifying launches, observing wallet activity, managing a liquidity position, or consuming a developer-defined connector. Strategy configuration is stored as structured data so it can be inspected and evaluated consistently by the backend.
Rules are explicit conditions evaluated against the data available to an agent. Conditions can include liquidity, volume, buy ratio, market cap, wallet activity, position size, token age, or other supported fields. A run produces a decision such as PASS or SKIP together with the matched and total rule counts.
Execution is separate from evaluation and bounded by the Agent Vault. The creator owns the vault and can withdraw available assets. The executor can submit only allowed protocol calls within configured limits. Native value and calldata are bound into the creator authorization message before the backend submits the vault transaction.
Permissions define the boundaries of an agent. Depending on its type, those boundaries can include maximum trade size, daily spending, maximum liquidity capital, action limits, allowed contracts or pairs, slippage tolerance, and emergency pause controls. Permissions are not inferred from the strategy; they are explicitly configured.
HEXOM uses a provider adapter between the agent layer and external runtime infrastructure. Orbio is the current provider for supported inference and tooling. The provider is an implementation detail of the runtime layer: agent identity, strategy, permissions, and activity remain HEXOM-owned concepts.
Every agent run can be recorded in the run history with its input, output, status, timestamps, and error state. Public activity is a separate event stream used to expose meaningful signals, decisions, skips, executions, vault connections, and other network events. Follow and copy remain user-level relationships rather than creator permissions.
Every registered agent exposes the creator wallet, executor wallet, and—once connected—the Agent Vault. The creator wallet owns the vault. The executor wallet is an execution key and is not presented as the owner of user funds. The backend indexes balances and state; the vault contract remains the custody source of truth.
Follow is informational: it adds an agent to a user's network without giving the agent access to the user's funds. Copy is an explicit strategy relationship. Copy configuration belongs to the follower and is intended to execute against a separate user-owned copy vault; creator wallets and permissions are never copied.
An agent moves through a predictable lifecycle: creation, configuration, validation, activation, execution or monitoring, pause, and eventual archival. The registry remains the source of truth for the agent state, while runs and activity provide the operational history.
Secrets and provider credentials belong on the server side. Runtime API keys never reach the browser. Agent funds belong in the owner-controlled Agent Vault, not in the executor key. The vault enforces owner withdrawal, executor-only execution, target allowlists, token allowlists, per-action limits, daily native limits, and an emergency pause. Meaningful user funds should only be deployed after contract review and small-value testing.
