Configuration¶

SteadyText can be configured through environment variables to customize its behavior for different use cases.

Environment Variables¶

Model Configuration¶

STEADYTEXT_MAX_CONTEXT_WINDOW: Maximum context window size (default: auto-detected per model)
STEADYTEXT_ALLOW_MODEL_DOWNLOADS: Allow automatic model downloads (default: true)
STEADYTEXT_DISABLE_DAEMON: Disable daemon mode (default: false)
STEADYTEXT_UNSAFE_MODE: Enable unsafe mode for remote models (default: false)

Remote Model Configuration (Unsafe Mode)¶

OPENAI_API_KEY: API key for OpenAI models (required for openai:* models)
CEREBRAS_API_KEY: API key for Cerebras models (required for cerebras:* models)

Cache Configuration¶

Generation Cache¶

STEADYTEXT_GENERATION_CACHE_CAPACITY: Maximum number of cache entries (default: 256)
STEADYTEXT_GENERATION_CACHE_MAX_SIZE_MB: Maximum cache file size in MB (default: 50.0)

Embedding Cache¶

STEADYTEXT_EMBEDDING_CACHE_CAPACITY: Maximum number of cache entries (default: 512)
STEADYTEXT_EMBEDDING_CACHE_MAX_SIZE_MB: Maximum cache file size in MB (default: 100.0)

Cache Backend Configuration¶

STEADYTEXT_CACHE_BACKEND: Cache backend type (default: sqlite)
sqlite: Local SQLite database
d1: Cloudflare D1 distributed database
memory: In-memory cache (ephemeral)

D1 Backend Configuration¶

STEADYTEXT_D1_API_URL: D1 API endpoint URL
STEADYTEXT_D1_API_KEY: D1 API authentication key
STEADYTEXT_D1_BATCH_SIZE: Batch size for D1 operations (default: 50)

Daemon Configuration¶

STEADYTEXT_DAEMON_HOST: Daemon host address (default: localhost)
STEADYTEXT_DAEMON_PORT: Daemon port (default: 5557)

Shell Integration Configuration¶

STEADYTEXT_SUGGEST_ENABLED: Enable shell suggestions (default: 1)
STEADYTEXT_SUGGEST_MODEL_SIZE: Model size for suggestions (default: small)
STEADYTEXT_SUGGEST_STRATEGY: Suggestion strategy (default: context)
STEADYTEXT_SUGGEST_ASYNC: Enable async suggestions (default: 1)

Configuration Examples¶

High-Performance Setup¶

export STEADYTEXT_MAX_CONTEXT_WINDOW=32768
export STEADYTEXT_GENERATION_CACHE_CAPACITY=1024
export STEADYTEXT_GENERATION_CACHE_MAX_SIZE_MB=200.0
export STEADYTEXT_EMBEDDING_CACHE_CAPACITY=2048
export STEADYTEXT_EMBEDDING_CACHE_MAX_SIZE_MB=500.0

Minimal Memory Setup¶

export STEADYTEXT_MAX_CONTEXT_WINDOW=2048
export STEADYTEXT_GENERATION_CACHE_CAPACITY=64
export STEADYTEXT_GENERATION_CACHE_MAX_SIZE_MB=10.0
export STEADYTEXT_EMBEDDING_CACHE_CAPACITY=128
export STEADYTEXT_EMBEDDING_CACHE_MAX_SIZE_MB=20.0

Distributed Cache Setup¶

export STEADYTEXT_CACHE_BACKEND=d1
export STEADYTEXT_D1_API_URL=https://your-worker.workers.dev
export STEADYTEXT_D1_API_KEY=your-api-key
export STEADYTEXT_D1_BATCH_SIZE=100

Testing Configuration¶

export STEADYTEXT_CACHE_BACKEND=memory
export STEADYTEXT_ALLOW_MODEL_DOWNLOADS=true
export STEADYTEXT_DISABLE_DAEMON=true

Remote Model Configuration (Unsafe Mode)¶

# Enable unsafe mode
export STEADYTEXT_UNSAFE_MODE=true

# Set API keys
export OPENAI_API_KEY=sk-your-openai-key
export CEREBRAS_API_KEY=your-cerebras-key

# Use remote models
python -c "import steadytext; print(steadytext.generate('Hello', model='openai:gpt-4o-mini'))"

Platform-Specific Configuration¶

Linux/macOS¶

Configuration files and caches are stored in: - Cache: ~/.cache/steadytext/ - Models: ~/.cache/steadytext/models/

Windows¶

Configuration files and caches are stored in: - Cache: %LOCALAPPDATA%\steadytext\steadytext\ - Models: %LOCALAPPDATA%\steadytext\steadytext\models\

Advanced Configuration¶

Custom Model Paths¶

You can specify custom model repositories and filenames:

import steadytext

# Use custom model repository
text = steadytext.generate(
    "Hello world",
    model_repo="ggml-org/gemma-3n-E2B-it-GGUF",
    model_filename="gemma-3n-E2B-it-Q8_0.gguf"
)

Context Window Management¶

import os
import steadytext

# Set maximum context window
os.environ["STEADYTEXT_MAX_CONTEXT_WINDOW"] = "8192"

# Generate with automatic context management
text = steadytext.generate("Your very long prompt here...")

Daemon Management¶

from steadytext.daemon.client import use_daemon

# Force daemon usage
with use_daemon():
    text = steadytext.generate("Hello world")

Troubleshooting Configuration¶

Common Issues¶

Models not downloading: Set STEADYTEXT_ALLOW_MODEL_DOWNLOADS=true
Cache growing too large: Reduce *_CACHE_MAX_SIZE_MB values
Memory usage high: Reduce *_CACHE_CAPACITY values
Daemon connection issues: Check STEADYTEXT_DAEMON_HOST and STEADYTEXT_DAEMON_PORT

Debug Configuration¶

export STEADYTEXT_DEBUG=1
export STEADYTEXT_VERBOSE=1

For more troubleshooting help, see the Troubleshooting Guide.