For AI Agents¶
datason is designed to be easily discoverable and usable by AI coding agents (Claude, GPT, Copilot, etc.).
Machine-Readable Documentation¶
datason provides llms.txt and llms-full.txt following the llms.txt standard.
llms.txt-- Short summary with links to API docs, examples, and source filesllms-full.txt-- Complete API reference with all function signatures, config options, and ready-to-use code examples in a single file
Why datason for AI Agents?¶
AI agents frequently need to serialize complex Python objects when:
- Generating data pipeline code -- ML model outputs contain NumPy arrays, timestamps, UUIDs
- Building API endpoints -- Response dicts contain datetime objects, Decimal prices
- Logging and debugging -- Mixed-type dicts need JSON output for structured logging
- Persisting state -- Agent state includes datetime, arrays, paths
With datason, agents can write:
Instead of:
import json
class CustomEncoder(json.JSONEncoder):
def default(self, obj):
if isinstance(obj, datetime):
return obj.isoformat()
if isinstance(obj, np.ndarray):
return obj.tolist()
if isinstance(obj, UUID):
return str(obj)
# ... 20 more type checks
return super().default(obj)
json.dumps(data, cls=CustomEncoder)
Quick Reference for Agents¶
Installation¶
pip install datason
pip install datason[numpy] # if using NumPy
pip install datason[pandas] # if using Pandas
pip install datason[ml] # if using PyTorch/TF/sklearn
API (5 functions)¶
import datason
datason.dumps(obj) # -> JSON string
datason.loads(s) # -> Python object (types reconstructed)
datason.dump(obj, file) # -> write JSON to file
datason.load(file) # -> read JSON from file
with datason.config(sort_keys=True):
datason.dumps(obj) # -> sorted JSON string