Skip to content

Releases: pollinations/search.elixpo

lix-open-cache v2.1.6

27 Mar 07:47
@Circuit-Overtime Circuit-Overtime
lix-open-cache
5480c84

Choose a tag to compare

View all tags
lix-open-cache v2.1.6 Latest
Latest

lix-open-cache v2.1.6

Standalone multi-layer Redis caching + Huffman disk archival for conversational AI.

Install

pip install lix-open-cache==2.1.6

Quick Start

from lix_open_cache import CacheConfig, CacheCoordinator

config = CacheConfig(redis_host="localhost", redis_port=6379)
cache = CacheCoordinator(session_id="user-abc", config=config)

# Store & retrieve conversation context
cache.add_message_to_context("user", "What's the weather in Tokyo?")
cache.add_message_to_context("assistant", "22°C and sunny.")
history = cache.get_context_messages()

# Semantic cache — skip LLM on similar queries
import numpy as np
embedding = np.random.rand(384).astype(np.float32)
cached = cache.get_semantic_response("https://weather.com", embedding)

Three Cache Layers

Layer Purpose Backend TTL
Session Context Window Rolling 20-message window + disk overflow Redis DB 2 + .huff files 24h
Semantic Query Cache Deduplicate similar queries (cosine ≥ 0.90) Redis DB 0 5 min
URL Embedding Cache Cache embedding vectors per URL Redis DB 1 24h

Key Features

  • Two-tier hybrid storage — Redis hot window + Huffman-compressed disk cold archive
  • LRU eviction daemon — auto-migrates idle sessions to disk, re-hydrates on return
  • smart_context() — recent messages + semantically relevant history from disk
  • Pure Python Huffman codec — ~54% compression, zero native dependencies
  • CacheConfig dataclass — all tunables in one place, 12-factor env var support
  • Connection pooling — shared Redis pools keyed by (host, port, db)

Dependencies

Only 3: redis, numpy, loguru

Links

Assets 4
Loading

lix-open-search v2.1.1

26 Mar 15:21
@Circuit-Overtime Circuit-Overtime
lix-open-search
62277f5

Choose a tag to compare

View all tags
lix-open-search v2.1.1

lix-open-search v2.1.1

Python client SDK for lixSearch — multi-tool AI search with web, video, image, and deep research.

Install

pip install lix-open-search==2.1.1

Quick Start

from lix_open_search import LixSearch

lix = LixSearch("http://localhost:9002")

# One-shot search
result = lix.search("quantum computing breakthroughs 2026")
print(result.content)

# Streaming
for chunk in lix.search_stream("latest AI papers"):
    print(chunk.content, end="", flush=True)

# Multi-turn conversation
result = lix.chat([
    {"role": "user", "content": "Compare Tesla and BYD sales"}
], session_id="my-session")

# Multimodal (text + image)
result = lix.search("What is this?", images=["https://example.com/photo.jpg"])

# Raw URLs without LLM
urls = lix.surf("best Python frameworks", limit=10)

Async

from lix_open_search import AsyncLixSearch

async with AsyncLixSearch("http://localhost:9002") as lix:
    result = await lix.search("SpaceX updates")
    async for chunk in lix.search_stream("AI papers"):
        print(chunk.content, end="", flush=True)

Features

  • Sync + Async clients (LixSearch / AsyncLixSearch)
  • Streaming with parsed StreamChunk objects
  • Multi-turn sessions with server-side memory
  • Multimodal — text + up to 3 images
  • Surf — raw URL/image search without LLM synthesis
  • OpenAI-compatible — also works with the standard OpenAI Python client
  • Single dependency — just httpx

Self-Host with Docker

docker pull elixpo/lixsearch
docker compose -f package/lix_open_search_pkg/docker-compose.yml up -d

Links

Loading