sensa-ai 0.2.0

Real-time environmental context for AI agents

🧠 sensa

Real-time environmental context for AI agents.

Your AI agent doesn't know what time it is. It doesn't know if it's raining, what Bitcoin costs, or what just happened in the news. Sensa fixes that — injecting a compact, token-efficient snapshot of the real world into every prompt.

Quick Install

pip install sensa

Quickstart

from sensa import SensaClient

client = SensaClient(channels=["time", "weather", "crypto", "news"], location="Austin, TX")
context = client.get_context_sync()
print(context)

Example output:

[SENSA — Tue Apr 7, 2026 3:15 PM CDT]
⏱ Session: 47m | Last call: 2m ago
🌤 Austin, TX: 78°F, partly cloudy, wind 8mph
📈 BTC: $94,230 (+2.1%) | ETH: $3,847 (+1.8%) | SOL: $142 (+0.5%)
📰 Fed signals rate hold | Tech earnings beat estimates | Oil prices steady

Channels

Channel	Data	API	Key Required
time	Date, time, session tracking	stdlib	No
weather	Temp, conditions, wind	wttr.in	No
crypto	BTC, ETH, SOL prices + 24h %	CoinGecko	No
news	Top 3 headlines	RSS (BBC, AP)	No

Configuration

client = SensaClient(
    channels=["time", "weather", "crypto"],
    location="Austin, TX",           # Used by weather channel
    timezone="America/Chicago",      # Timezone for time channel
    max_tokens=200,                  # Token budget for output
    api_keys={                       # Optional premium API keys
        "openweathermap": "your-key",
    },
    staleness_thresholds={           # Minutes before data is flagged stale
        "weather": 15,
        "crypto": 5,
        "news": 30,
    },
)

Async Usage

import asyncio
from sensa import SensaClient

async def main():
    client = SensaClient(channels=["time", "weather"])
    context = await client.get_context()
    print(context)

asyncio.run(main())

Zero Config

Sensa works with zero configuration. Just SensaClient() gives you the time channel out of the box.

How It Works

1. Channels fetch data from free public APIs (no keys needed for basic use)
2. Compression fits all channel output within your token budget using priority-based allocation
3. Staleness detection flags data that's too old, so your agent knows what to trust

Contributing

Contributions welcome! Please open an issue or PR on GitHub.

1. Fork the repo
2. Create a feature branch
3. Add tests for new functionality
4. Submit a pull request

License

MIT — see LICENSE for details.