Skip to content
Provara

Configuration

Overview

Section titled “Overview”

Provara uses a layered configuration system powered by pydantic-settings:

  1. Environment variables (highest priority)
  2. .env file in the project root
  3. Default values (lowest priority)

All environment variables use the AGENT_HUB_ prefix.

Configuration Reference

Section titled “Configuration Reference”

Server Settings

Section titled “Server Settings”
VariableDefaultDescription
AGENT_HUB_HOST127.0.0.1API bind address
AGENT_HUB_PORT8787API port (1-65535)
AGENT_HUB_BASEhttp://127.0.0.1:8787API base URL (used by clients)

Authentication

Section titled “Authentication”
VariableDefaultDescription
AGENT_HUB_TOKEN(empty)API authentication token

The token can also be loaded from a file at runtime/agent_hub_token.txt. The file-based approach is preferred for production deployments.

Paths

Section titled “Paths”
VariableDefaultDescription
AGENT_HUB_ROOT(auto-detect)Project root directory
AGENT_HUB_RUNTIME_DIRruntimeRuntime data directory (relative to root)
AGENT_HUB_WORKSPACE_DIRworkspaceAgent workspace directory (relative to root)

The project root is auto-detected by walking up from the source file to find pyproject.toml. Override with AGENT_HUB_ROOT if running from a different directory.

Logging

Section titled “Logging”
VariableDefaultDescription
AGENT_HUB_LOG_LEVELINFOLogging verbosity

Supported levels: DEBUG, INFO, WARNING, ERROR, CRITICAL

Rate Limiting

Section titled “Rate Limiting”
VariableDefaultDescription
AGENT_HUB_RATE_LIMIT_REQUESTS100Max requests per window
AGENT_HUB_RATE_LIMIT_WINDOW_SECONDS60Window duration in seconds

Rate limiting is applied per client IP address.

Security

Section titled “Security”
VariableDefaultDescription
AGENT_HUB_STRICT_MODE0Set to 1 to deny all unknown commands

Optional Integrations

Section titled “Optional Integrations”
VariableDefaultDescription
OLLAMA_BASE_URLhttp://127.0.0.1:11434Ollama API base URL
OLLAMA_MODELllama3.1:8bDefault Ollama model

Example .env File

Section titled “Example .env File”
Terminal window
# Server
AGENT_HUB_HOST=127.0.0.1
AGENT_HUB_PORT=8787
AGENT_HUB_BASE=http://127.0.0.1:8787


# Authentication
AGENT_HUB_TOKEN=your-secure-token-here


# Logging
AGENT_HUB_LOG_LEVEL=INFO


# Rate Limiting
AGENT_HUB_RATE_LIMIT_REQUESTS=100
AGENT_HUB_RATE_LIMIT_WINDOW_SECONDS=60


# Security
AGENT_HUB_STRICT_MODE=0

Configuration Class

Section titled “Configuration Class”

Under the hood, configuration is managed by a Pydantic Settings class in src/server/config.py:

from pydantic_settings import BaseSettings


class Settings(BaseSettings):
    model_config = SettingsConfigDict(
        env_prefix="AGENT_HUB_",
        env_file=".env",
    )


    host: str = "127.0.0.1"
    port: int = 8787
    token: str = ""
    log_level: Literal["DEBUG", "INFO", "WARNING", "ERROR", "CRITICAL"] = "INFO"
    strict_mode: bool = False
    # ... more fields

Settings are cached via @lru_cache and can be reloaded by calling get_settings.cache_clear().

Runtime Directories

Section titled “Runtime Directories”

Provara automatically creates the following directory structure on startup:

runtime/
├── pending/         # Commands awaiting approval (JSON files)
├── denied/          # Denied commands archive
├── runs/            # Execution history (one dir per run)
│   └── {run_id}/
│       └── record.json
├── audit.log        # Append-only audit trail
└── agent_hub_token.txt  # Token file (optional)


workspace/
├── inbox/           # Script staging area
├── outbox/          # Output files
└── logs/            # Agent logs

Next Steps

Section titled “Next Steps”