---
title: chat, eval, verify
description: Interactive TUI chat session, benchmarking harness, and screenshot verification.
sidebar:
  order: 6
  label: chat, eval, verify
---

## `chat`, `eval`, and `verify`

Interactive terminal interfaces, seeded benchmarks, and visual verification.

---

## `niki chat`

Launches a terminal user interface (TUI) powered by `ratatui` for interactive pair programming:

```bash
niki chat
```

### Supported Slash Commands
* `/compact`: Compact session history to save token context.
* `/model`: Switch models or providers on the fly.
* `/cost`: Inspect session token usage and estimated spend.
* `/memory store <fact>`: Persist context into project memory.
* `/undo` / `/redo`: Revert or reapply previous agent steps.
* `/export`: Export session transcript to markdown.

---

## `niki eval`

Runs NIKI against standardized or custom defect datasets to benchmark accuracy and revision efficiency:

```bash
# Run evaluation harness against dataset
niki eval --dataset evals/dataset.toml --limit 5

# Output machine-readable JSON metrics
niki eval --dataset evals/dataset.toml --format json --out results.json
```

---

## `niki verify`

Performs visual UI regression testing by comparing rendered screenshots against baselines:

```bash
niki verify "Check that the navbar aligns properly on mobile screens" --baseline assets/baseline.png
```