seval

AI-powered security research CLI built in Rust. Features a split-pane TUI dashboard, agentic tool execution via any LLM (AWS Bedrock or OpenRouter), session persistence, and context compression.

Features

Streaming chat with any LLM through AWS Bedrock or OpenRouter (Claude, GPT, Llama, Gemini, etc.)
10 built-in tools the AI can invoke autonomously (shell, file ops, search, web)
Split-pane TUI powered by ratatui -- chat on the left, tool output on the right
Session management with SQLite persistence, resume/export/import
Context compression via tiktoken to stay within model limits
Approval modes to control tool execution safety
Project memories the AI can save and recall across sessions
Configurable deny rules to block dangerous shell commands
Markdown rendering with syntax highlighting in the terminal
System prompt override via ~/.seval/system.md

Prerequisites

Rust toolchain (edition 2024, so Rust 1.85+)
AWS credentials with Bedrock access, or an OpenRouter API key
Brave Search API key (optional, enables the web_search tool)

Setup

Run seval for the first time or use seval init to launch the interactive configuration wizard. This creates ~/.seval/config.toml.

To re-run the wizard and overwrite an existing config:

seval init --force

Configuration files

File	Purpose
`~/.seval/config.toml`	Global config (credentials, provider, model, approval mode, deny rules)
`.seval/config.toml`	Project-local overrides (approval mode, deny rules, AWS settings)
`~/.seval/system.md`	Custom system prompt (replaces the default)

Example global config

[provider]
active = "bedrock"   # or "open-router"

[bedrock]
access_key_id = "AKIA..."
secret_access_key = "..."
region = "us-east-1"

[openrouter]
api_key = "sk-or-..."

[tools]
approval_mode = "default"
max_turns = 25
deny_rules = [
  "rm -rf /",
  "rm -rf /*",
  "chmod 777 /",
  "mkfs.*",
  "> /dev/sd*",
  "dd if=* of=/dev/*",
]

brave_api_key = "BSA..."

Build & Install

cargo build --release
# Binary at target/release/seval

Or run directly:

cargo run --release

Usage

seval [OPTIONS] [COMMAND]

Commands:
  init  Initialize configuration (interactive wizard)

Options:
  --profile <PROFILE>          AWS profile name
  --region <REGION>            AWS region
  --model <MODEL>              Bedrock model ID
  --approval-mode <MODE>       Tool approval mode [plan|default|auto-edit|yolo]
  --config <PATH>              Path to configuration file
  -h, --help                   Print help
  -V, --version                Print version

Approval modes

Mode	Behavior
`plan`	Read-only, no tool execution
`default`	Ask before write operations
`auto-edit`	Auto-approve file edits, ask for shell commands
`yolo`	Approve everything automatically

Slash commands

Type these in the chat input:

Command	Description
`/model [name]`	Switch AI model (show current if no name)
`/sessions`	List saved sessions
`/sessions resume <id>`	Resume a saved session
`/sessions delete <id>`	Delete a saved session
`/import <path>`	Import a session from JSON
`/export [id]`	Export session to JSON
`/memory`	List project memories
`/memory delete <id>`	Delete a memory entry
`/help`	Show help
`/clear`	Clear conversation history
`/quit` or `/q`	Quit

Tools

The AI has access to 10 built-in tools during agentic execution:

Tool	Description
`shell`	Execute shell commands
`read`	Read file contents
`write`	Write/create files
`edit`	Apply targeted edits to files
`grep`	Search file contents with regex
`glob`	Find files by glob pattern
`ls`	List directory contents
`web_fetch`	Fetch and extract text from a URL
`web_search`	Search the web via Brave Search API
`save_memory`	Persist a memory for future sessions

TUI Layout

The interface is a split-pane terminal UI:

Left pane -- Chat conversation with streaming markdown rendering
Right pane -- Tool execution output and approval prompts
Bottom bar -- Input area and status indicators (model, token count, session)

License

This project is licensed under the GNU Affero General Public License v3.0.

seval 0.1.2