Agent Self-Service (pando_setup)

The pando_setup tool lets the AI agent inspect Pando’s configuration, discover available models, check session usage, and activate slash commands — all without leaving the conversation. It’s like giving the agent its own control panel.

This tool is always available to the agent and requires no configuration. It’s read-only for configuration and providers, and can activate certain slash commands when appropriate.

How It Works

The agent uses pando_setup automatically during conversations. You don’t call it directly — the AI decides when it needs to check settings, find available models, or run a command.

What the Agent Can Do

Command	Purpose
`help`	List available commands or get usage for a specific one
`config`	Read the active configuration (read-only, same view as TUI/WebUI settings)
`providers`	List configured provider accounts with types, credentials, and model counts
`models`	Browse available models with context window, pricing, and capabilities
`session`	Check last turn’s token usage and accumulated session cost
`commands`	List available slash commands
`run <command>`	Activate a slash command for the current session

Examples

The agent might use these commands during a conversation:

Checking models: pando_setup models --provider copilot --detail — shows available Copilot models with pricing and context windows
Listing providers: pando_setup providers — shows all configured provider accounts
Viewing config: pando_setup config --search auth — searches configuration for auth-related settings
Running a command: pando_setup run /caveman lite — enables caveman mode for the session
Session stats: pando_setup session — shows token usage and cost for the current session

Model Discovery

The models command uses the live model registry, enriched with data from models.dev. It shows:

Canonical ID: e.g., copilot.gpt-5.4, anthropic.claude-sonnet-4-5
Context window: Token limit (e.g., 200K, 1M)
Pricing: Per-million-token input/output costs
Capabilities: Reasoning, image support, etc.
Knowledge cutoff: Training data cutoff date

Filter by provider, search by name, or limit results:

pando_setup models --provider anthropic --detail --limit 5

Configuration Access

The config command reads the same configuration that TUI and WebUI settings panels show. Secrets are masked (only last 4 characters visible) so the agent can tell “configured” from “unset” without seeing actual values.

pando_setup config                           # show all config
pando_setup config --search token            # search for token-related settings
pando_setup config TokenOptimization         # show a specific section

Slash Commands

The agent can activate certain slash commands via pando_setup run:

Mode commands (mutate session state):

/caveman, /caveman-finish
/ponytail, /ponytail-finish
/superpowers, /superpowers-finish
/learning, /learning-finish

Instruction commands (return instructions for the current turn):

/improve-agents-md
/vulnhunt, /vulnhunter-fix, /vulnhunt-fix-verify
Custom user: and project: commands

Some commands are blocked because they require surface-level interaction:

/goal* commands
/compact, /db-compact
All -finish closing commands

Security

Configuration is strictly read-only — no write path exists
Secrets are masked by key suffix (apikey, token, password, secret, etc.)
The tool description stays under 600 characters to keep token cost low
The tool is always included in the agent’s toolset (cannot be trimmed by context optimization)

Configuration Reference

The pando_setup tool requires no configuration. It is always available.

Setting	Description
None	The tool is built-in and always active

The pando_setup tool is designed to be cheap in tokens. Its description is minimal, and it discovers detail on demand through help and --help rather than enumerating everything upfront.

Learning Mode Configuration File Discovery