langstate 0.1.0

Scaffold-aware context compression for OpenAI-format messages — preserve state at 50%+ token reduction via local Ollama, OpenAI, or Anthropic

langstate

Scaffold-aware context compression for OpenAI-format messages. compress(messages) preserves all conversational state — facts, decisions, task status, user preferences — while reducing tokens by 50-54%.

Backed by the LPCI thesis: stateless LLMs hold state via language scaffold with transfer entropy approximately zero (Markov property). This is the productization of that proof.

What it does

from langstate import compress

compressed = compress(messages)
# Drop-in replacement: same OpenAI format, fewer tokens, state preserved
response = client.chat.completions.create(messages=compressed, model="gpt-4o")

The output is a valid OpenAI-format messages list:

• System prompts kept verbatim
• Last 4 turn-pairs kept verbatim
• Older turns compressed into a [SCAFFOLD STATE] system message via local or cloud model

Install

# From GitHub (PyPI coming soon):
pip install git+https://github.com/hermes-labs-ai/langstate.git

Requirements: Python 3.10+, no heavy dependencies (stdlib only). For local summarization, run Ollama locally:

ollama pull qwen3:4b

Adapters

langstate ships three built-in summarizer backends:

Adapter	Model	Cost	Key
local (default)	qwen3:14b via Ollama	zero	none
openai	gpt-4o-mini	API	OPENAI_API_KEY
anthropic	claude-opus-4-7	API	ANTHROPIC_API_KEY

from langstate import compress
from langstate.adapters import build

# Local Ollama (default, zero cost)
compressed = compress(messages)

# OpenAI
compressed = compress(messages, summarizer=build("openai"))

# Anthropic
compressed = compress(messages, summarizer=build("anthropic"))

# Any callable (prompt: str) -> str
compressed = compress(messages, summarizer=my_summarizer)

Configuration

compress(
    messages,
    preserve_recent=4,         # turn-pairs to keep verbatim (default: 4)
    min_turns_to_compress=6,   # skip compression for short conversations (default: 6)
    model="qwen3:4b",          # Ollama model when no summarizer is given
    summarizer=None,           # custom callable: (prompt: str) -> str
)

Adapter probe

from langstate.adapters import probe, REGISTRY

for name in REGISTRY:
    print(probe(name))
# {"name": "local", "available": True, "latency_ms": 423, ...}
# {"name": "openai", "available": False, "reason": "OPENAI_API_KEY not set", ...}

License

Apache-2.0

---

About Hermes Labs

Hermes Labs builds AI audit infrastructure for teams deploying AI agents in regulated environments. All tools are released as open-source software — MIT or Apache-2.0, no SaaS tier. The audit work is paid; the code is not.

hermes-labs.ai

OSS audit stack

Layer	Tool	Description
Static audit	lintlang	Agent-config static lint (HERM + H1-H7)
Static audit	rule-audit	Rule-logic audit: contradictions + gaps
Static audit	scaffold-lint	Scaffold budget + technique stacking
Static audit	intent-verify	Spec-drift checks
Runtime observability	little-canary	Prompt injection detection
Runtime observability	suy-sideguy	Runtime policy guard
Runtime observability	colony-probe	Prompt confidentiality audit
Regression & scoring	hermes-jailbench	Jailbreak regression benchmark
Regression & scoring	agent-convergence-scorer	N-agent output consistency
Supporting infra	claude-router	Model-tier + scaffold router
Supporting infra	quickthink	Compressed planning scaffold for local LLMs
Supporting infra	langstate	Scaffold-aware context compression
Supporting infra	agent-gorgon	Tool-fabrication defense for Claude Code
Supporting infra	zer0dex	Dual-layer agent memory
Supporting infra	forgetted	Mid-conversation incognito
Dev tools	repo-audit	Launch-readiness auditor
Dev tools	quick-gate-python	Python quality gate
Dev tools	quick-gate-js	JS/TS quality gate
Dev tools	csv-quality-gate	CSV preflight validation