Introduction

Agento is an enterprise agentic-coding CLI that pairs a fast, interactive OpenTUI terminal interface with a deterministic execution harness.

The Problem with Traditional Coding Agents

Most AI coding assistants operate on unverified trust:

1. Unchecked Mutations: The model directly writes files without verifying repository invariants, introducing regressions, breaking imports, or leaking secrets.
2. Hallucination Drift: When an agent attempts complex multi-step edits without upfront planning, intermediate broken states accumulate until the model loses context.
3. Slow or Missing Feedback: Slow test suites and linters either stall agent turns or are skipped entirely, leaving broken code behind.
4. Lack of Invariant Enforcement: Organizations cannot enforce strict coding standards (such as banned dependencies, required type safety, or protected paths) at the agent boundary.

The Agento Solution: Governed Agency

Agento flips the relationship between the AI model and the repository:

• The Harness Owns State: The AI only ever proposes edits, todo updates, and commands. State transitions only occur when deterministic gates verify them.
• Plan Mode vs Act Mode: Architecture and planning are strictly separated from execution. In Plan Mode, the agent explores and outlines tasks. Act Mode work requires an approved plan before any file can be modified.
• LINTOs (Pre-Write Guardrails): Repository rules are evaluated directly against in-memory edit proposals. If a rule is violated, the edit never touches disk.
• Checks (Post-Write Feedback): Fast compiler and linter commands run immediately after an edit to provide instant diagnostic feedback, allowing the model to self-correct without human intervention.
• Milestone Verification: Completing a task or milestone triggers comprehensive verification tests to guarantee repository integrity before proceeding.

Polyglot by Construction

Agento is language-agnostic. It does not hardcode language-specific assumptions or rely on brittle heuristics. Verification commands and linters are discovered dynamically from your project configuration (such as lockfiles, build scripts, and tool configs), ensuring compatibility across TypeScript, Rust, Go, Python, C++, and more.