LQCC 0.7.1

Lightweight Queryable Context Compression is a local-first .capsule context dictionary for long-running AI conversations.

LQCC saves the full visible history locally, extracts a searchable context dictionary, and gives the next AI session only the small context packet it needs.

What works in 0.7.1

• no-SQL packed .capsule file format
• Python CLI for Linux, Windows, and macOS
• one-command beginner flow: lqcc quick chat.md
• interactive terminal menu: lqcc or lqcc menu
• local HTTP daemon for automatic appends and retrieval
• OpenAI-compatible non-streaming proxy that records requests and replies
• command wrapper that records command output as tool context
• search and resume from the capsule without loading full history
• PDF, image, text, code, and binary attachments with AI-readable sidecars
• full-history export and capsule verification
• reader skill for Codex, Claude Code, Cursor, and other agents

Install

Test install via TestPyPI:

From TestPyPI:

pip install --index-url https://test.pypi.org/simple/ --extra-index-url https://pypi.org/simple lqcc

From source:

git clone https://github.com/zjnbwxq/lqcc.git
cd lqcc
python -m pip install -e ".[multimodal]"

Check:

lqcc --version

Expected:

lqcc 0.7.1

Fastest use

lqcc quick examples/demo_chat.md

This creates examples/demo_chat.capsule and prints a small packet you can paste into a new AI chat.

Normal manual flow

lqcc build chat.md -o project.capsule --title "My Project"
lqcc search project.capsule "what did we decide?"
lqcc resume project.capsule --task "continue the project" --budget 800
lqcc verify project.capsule

Interactive menu

lqcc

or:

lqcc menu

The menu lets users build, resume, search, append, attach, verify, start a daemon, or start a proxy without remembering command syntax.

Local daemon

Use the daemon when another tool or agent should write into and read from a capsule automatically.

lqcc daemon project.capsule --port 8765

Endpoints:

GET  /health
GET  /stats
POST /append
POST /append-many
POST /resume
POST /search
POST /get
POST /attach

Example:

curl -X POST http://127.0.0.1:8765/append \
  -H "Content-Type: application/json" \
  -d '{"role":"user","content":"Decision: keep active context small."}'

OpenAI-compatible proxy

Use the proxy when an API client should send messages through LQCC so visible messages are automatically written to the capsule.

export OPENAI_API_KEY="your-key"
lqcc proxy project.capsule \
  --upstream https://api.openai.com/v1/chat/completions \
  --context-mode auto \
  --port 8765

Then point the client to:

http://127.0.0.1:8765/v1/chat/completions

Context modes:

pass    record messages, forward original request
resume  forward a capsule resume packet plus recent messages
auto    use pass until the request becomes large, then use resume

Current limitation: streaming proxy responses are not supported in 0.7.1. Use non-streaming requests.

Command wrapper

Use wrap to record command output as tool context:

lqcc wrap project.capsule -- python -m pytest

This records the command start, exit code, stdout, and stderr into the capsule.

Agent reader skill

See:

reader-skill/SKILL.md
reader-skill/SKILL.zh-CN.md

The skill tells an agent to use lqcc resume first, then lqcc search or lqcc get only when more evidence is needed.

Scope

LQCC 0.7.1 handles visible text and file artifacts. It does not capture hidden chain-of-thought. It does not include a browser extension yet. Browser support, desktop UI, stronger multimodal indexing, and stable v1.0 format guarantees are future work.

My Links

https://linux.do/u/zjnbwxq

License

MIT.