Semantic quantization for multi-agent AI coordination.
Slipstream encodes coordination messages as compact Force+Object intents:
SLIP v3 <src> <dst> <Force> <Object> [payload...]
Typical JSON coordination messages (~40+ tokens) compress to ~6-8 wire tokens.
- Quick adoption guide
- LangGraph integration guide
- SDK reference
- v3.0 -> v3.1 migration notes
- Protocol invariants
- Canonical paper source (v3.1 markdown)
pip install slipcorefrom slipcore import format_slip, parse_slip, quantize, render_human
# Direct formatting
wire = format_slip("alice", "bob", "Request", "Review", ["auth"])
assert wire == "SLIP v3 alice bob Request Review auth"
# Think -> Quantize -> Transmit helper
wire2 = quantize(
"Please review the authentication code",
src="dev",
dst="reviewer",
)
msg = parse_slip(wire)
print(msg.force, msg.obj, msg.payload)
print(render_human(msg))Fallback messages are now strict:
Fallbackmessages must include a ref token.- Ref token must be alphanumeric and 1-16 chars.
- Use
parse_slip_legacy()only for explicit migration of old permissive wires.
from slipcore import format_fallback, parse_slip_legacy
wire = format_fallback("qa", "planner", "ref7f3a")
legacy = parse_slip_legacy("SLIP v3 qa planner Fallback Generic")
assert legacy.fallback_ref == "reflegacy"Observe, Inform, Ask, Request, Propose, Commit, Eval, Meta, Accept, Reject, Error, Fallback
- Model: anthonym21/slipstream-glm-z1-9b
- Dataset: anthonym21/slipstream-tqt
- A2A transport extension: extensions/a2a-slipstream/v1
@misc{maio2025slipstream,
title={Slipstream: Semantic Quantization for Efficient Multi-Agent Coordination},
author={Maio, Anthony},
year={2025},
url={https://github.com/anthony-maio/slipcore}
}Apache 2.0.