An all-in-one AI ecosystem browser in the terminal — explore models, benchmarks, coding agents, and provider status via TUI/CLI
models (crate name modelsdev) is a Rust-native terminal tool that aggregates AI ecosystem information from models.dev, Artificial Analysis, GitHub Releases, and various status pages into a single interface via dual TUI/CLI modes. Its four core modules include:
Model Browsing: 4,000+–4,200+ models and 85+–114+ providers (data from models.dev by SST team), three-column layout showing model list, details, and metadata, filtering by capability tags/price/context window, sorting by name/date/cost/context, cross-provider search, clipboard copy, and RTFO (real-time streaming first-token latency) capability indicator.
Coding Agent Tracking: 12+ built-in agents (Claude Code, Gemini CLI, Codex, etc.), automatic version detection via GitHub Releases, styled changelog browsing (comrak parsing → terminimad terminal rendering), real-time service health monitoring, custom agents and aliases via config.toml.
Benchmark Comparison: ~483+ entries from Artificial Analysis, metrics covering quality index, inference speed, and pricing, with head-to-head comparison tables, scatter plots, and radar charts. Benchmark data auto-updated daily via GitHub Actions.
Status Monitoring: Real-time health monitoring of 23+ AI providers across 7 status page platforms (Statuspage, BetterStack, Instatus, incident.io, etc.), with a dashboard featuring health gauges, incident cards, and maintenance cards.
Interaction Modes: TUI built on ratatui 0.30 + crossterm 0.29 with arrow key navigation, [/] tab switching, / search, ? context help, and Jaro-Winkler fuzzy matching; CLI built on clap 4 (derive mode) with full subcommand set (models list/show/search/providers, models benchmarks list/show, models status list/show, agents status/latest), JSON output, interactive selectors (dialoguer fuzzy-select), and bash/zsh/fish shell completions.
Installation: Homebrew (brew install models), Cargo (cargo install modelsdev), Scoop (scoop install extras/models), AUR (paru -S models-bin), prebuilt binaries and .deb/.rpm packages from GitHub Releases. Release builds use strip + LTO + codegen-units=1 + panic=abort optimizations.
Scope: The tool does not perform model inference or agent execution — it focuses on three information-consumption scenarios: browsing, comparison, and monitoring. Data depends on external APIs; offline capabilities are limited.