OpenPraxis

An OpenClaw-native knowledge retention skill that turns raw inputs into structured practice so you can use what you know, not just store it.

OpenClaw users: zero extra API key configuration — the host agent's model is reused automatically. Standalone CLI users: configure your own provider with praxis llm setup.

Requirements

• Python 3.11+

OpenClaw Users: Zero-Config Path

When running as an OpenClaw skill, OpenPraxis delegates all LLM calls to the host agent. No local API key is needed.

pip install openpraxis
export OPENPRAXIS_MODE=openclaw

Then use the skill directly from your OpenClaw agent — import knowledge, generate practice, evaluate answers, and produce insight cards.

See openclaw-knowledge-coach/SKILL.md for the full skill workflow.

Standalone CLI Users: Manual Setup

For use outside of OpenClaw, configure your own provider and API key:

pip install openpraxis
praxis llm setup
praxis llm show

Or install from source (for development):

git clone https://github.com/Sibo-Zhao/OpenPraxis.git
cd OpenPraxis
pip install -e ".[dev]"

Supported providers:

• openai (default): native structured output parse
• doubao: native structured output parse
• kimi / deepseek: JSON mode + Pydantic validation

API key env vars (higher priority than config file):

• OPENAI_API_KEY for openai
• ARK_API_KEY for doubao
• MOONSHOT_API_KEY for kimi
• DEEPSEEK_API_KEY for deepseek

Usage

praxis add <file> [--type report|interview|reflection|idea]
praxis practice <input_id>
praxis answer <scene_id> [--editor] [--file <path>]
praxis insight [<input_id>] [--type <insight_type>] [--min-intensity <n>]
praxis show <id>
praxis export [--format md|json] [--output <path>]
praxis list [--type report|interview|reflection|idea] [--limit N]

praxis add accepts both text/markdown files and common image formats (.png, .jpg, .webp, ...). For images, OpenPraxis uses a vision-capable model to extract readable text first (providers: openai or doubao).

Global runtime LLM overrides (for a single command, standalone CLI mode):

praxis --provider doubao --model doubao-seed-1-6-251015 add note.md
praxis --provider kimi --model kimi-k2-turbo-preview practice <input_id>
praxis --provider deepseek --model deepseek-chat answer <scene_id> --file answer.md

Architecture: Host-Managed LLM vs CLI-Managed LLM

OpenPraxis separates business workflow (schema, prompts, graph, scoring, persistence) from LLM call implementation via a backend abstraction:

• OpenClaw mode (OPENPRAXIS_MODE=openclaw): LLM calls are delegated to the host agent. No local API key needed.
• Standalone CLI mode (default): LLM calls use the locally configured provider and API key.

Nodes and graph logic depend only on LLMBackend interface, not on any specific provider. See ARCHITECTURE.md for details.

Development

pytest
ruff check src tests

Vision

Increase your "AI bandwidth" by converting fragmented inputs into reusable practice loops that build real transfer: faster recall, clearer decisions, better on-the-job application.

License

See the project repository.