Models
List, describe, and validate available LLM models from the CLI
praisonai models browses every model your install knows about — provider, capabilities, context window, and cost — so you pick the right one before you run.
Quick Start
- List all models
bash praisonai models list - Filter by provider or name
bash praisonai models list --provider openai praisonai models list gpt - Inspect a specific model
bash praisonai models describe gpt-4o - Validate a model ID before using it
bash praisonai models validate gpt-4o
Agent-centric example
from praisonaiagents import Agent
# `praisonai models describe gpt-4o` confirmed this model supports vision
agent = Agent(
name="Vision Agent",
instructions="Describe what's in the image.",
llm="gpt-4o",
)
agent.start("Describe this screenshot.")Subcommands
| Command | Purpose |
|---|---|
models list [SEARCH] [--provider P] [--json] | List available models grouped by provider |
models describe <model> | Full capabilities, limits, cost, and notes for one model |
models validate <model> | Check whether an ID is valid; suggests alternatives on a miss |
Flags
praisonai models list
| Flag | Type | Default | Description |
|---|---|---|---|
search (positional) | str | None | Substring filter on model ID |
--provider / -p | str | None | Filter to one provider |
--json | bool | False | Emit JSON instead of a Rich table |
praisonai models describe <model>
| Flag | Type | Default | Description |
|---|---|---|---|
model (positional) | str | — | Model ID (e.g. gpt-4o) |
praisonai models validate <model>
| Flag | Type | Default | Description |
|---|---|---|---|
model (positional) | str | — | Model ID to validate |
Fallback behaviour
If litellm is not installed, the CLI falls back to a curated static list covering OpenAI, Anthropic, Google, Groq, and Ollama models. All three subcommands work against the static list. Install litellm for the full live catalogue:
pip install 'praisonai[litellm]'Windows & non-UTF-8 terminals
The Capabilities column and ✅/❌ markers auto-switch to plain-text labels (tools, vision, reasoning, yes, no) when stdout can’t encode emoji — Windows PowerShell on the default cp1252 code page, redirected pipes, and some SSH sessions. Behaviour on macOS/Linux UTF-8 terminals is unchanged. Use --json for a byte-stable format in scripts and CI.
Best Practices
Validate model IDs before running agents
Catch typos in llm= before the agent fails on the first API call — praisonai models validate <id> exits 1 and suggests similar IDs on a miss.
Use –json for scripting
Pipe praisonai models list --json into jq to filter by capability or cost in CI pipelines.
Install litellm for the full catalogue
The static fallback covers only a handful of common models. Install praisonai[litellm] to unlock 100+ models with live pricing data.
On Windows PowerShell the table shows `tools` instead of `🔧 tools` — that's expected
The CLI probes sys.stdout.encoding and downgrades emoji capabilities to ASCII labels when the console can’t render them. To force emoji, switch your session to UTF-8 (chcp 65001) or run inside Windows Terminal / VS Code. To bypass rendering entirely, use praisonai models list --json.