Niyatna Docs

Bounded Agent Lifecycles

How Niyatna Agents execute tasks under sandbox constraints and generate structured proof.


The Agent Workflow is the execution cycle where specialized Niyatna Agents process goals. Agents do not run in black boxes; they operate under strict sandbox constraints and must produce verifiable proof before changes are committed.

Niyatna agent workflow

Agent Lifecycle Phases

  1. Intent Parsing: The agent reads the OpenIntent packet, maps the goal, and registers required tools, directories, and memory scopes.
  2. Execution: The agent performs operations (editing code, compiling scripts, checking route outputs) inside its isolated workspace container.
  3. Self-Verification: The agent compiles and tests the changes, generating the Proof of Intent bundle.
  4. Staging Gate: The agent pauses, staging all proposed changes, and flags the Command Room for human approval.

Bounded Autonomy

Niyatna Agents are equipped with advanced tools, wallets, and skills, but they remain a delegated workforce. The human operator retains final authority, maintaining veto control over all outbound actions.