Configuration¶

How to configure Perspt for your workflow.

Configuration Sources¶

Perspt loads configuration from (highest priority first):

CLI Arguments — perspt --model gpt-5.2
Environment Variables — export OPENAI_API_KEY=...
Config File — ~/.perspt/config.toml
Defaults — Built-in fallbacks

Config File Location¶

~/.perspt/config.toml

Create it:

mkdir -p ~/.perspt
perspt config --edit

Config File Format¶

# ~/.perspt/config.toml

[default]
provider = "openai"
model = "gpt-5.2"

[providers.openai]
api_key = "sk-..."

[providers.anthropic]
api_key = "sk-ant-..."

[agent]
architect_model = "gpt-5.2"
actuator_model = "claude-opus-4.5"
verifier_model = "gemini-3-pro"
energy_weights = [1.0, 0.5, 2.0]
stability_threshold = 0.1
max_retries_compile = 3
max_retries_tool = 5

Environment Variables¶

Variable	Description
`OPENAI_API_KEY`	OpenAI API key
`ANTHROPIC_API_KEY`	Anthropic API key
`GEMINI_API_KEY`	Google Gemini API key
`GROQ_API_KEY`	Groq API key
`COHERE_API_KEY`	Cohere API key
`XAI_API_KEY`	XAI (Grok) API key
`DEEPSEEK_API_KEY`	DeepSeek API key

CLI Configuration Commands¶

# Show current configuration
perspt config --show

# Set a value
perspt config --set default.model=gpt-5.2

# Edit in $EDITOR
perspt config --edit

Project Configuration¶

Initialize project-specific config:

cd my-project
perspt init --memory --rules

This creates:

my-project/
├── PERSPT.md         # Project memory/context
└── .perspt/
    ├── config.toml   # Project config
    └── rules.star    # Policy rules

PERSPT.md¶

Project memory file that provides context to the agent:

# My Project

## Overview
This is a Python web application using FastAPI.

## Architecture
- `api/` - REST endpoints
- `core/` - Business logic
- `tests/` - pytest suite

## Conventions
- Use type hints everywhere
- 100% test coverage required

Per-Session Configuration¶

Override for a single session:

perspt chat --model claude-opus-4.5

perspt agent \
  --architect-model gpt-5.2 \
  --actuator-model claude-opus-4.5 \
  "Create module"