The Serverless Working Memory for AI Agents. Powered by TiDB Cloud Zero.
Just like humans use "Working Memory" to process tasks, solve problems, and hold temporary context, AI Agents need a place to:
- Verify Code: Run SQL in a sandbox before outputting it.
- Process Data: Clean and analyze CSVs using SQL power.
- Hold Context: Store logs, RAG vectors, and state across sessions.
This repo provides standardized "Skills" that give your Agent this capability instantly.
A collection of self-contained skills. Each skill provisions a serverless TiDB database on-the-fly.
verify-code: The Agent runs generated SQL in a disposable DB to fix syntax errors before they reach the user. (Status: Ready)
data-refinery: Load messy CSVs -> SQL Cleaning -> Clean Export.data-diff: Compare two datasets using SQL set operations.
knowledge-vault: Instant Vector Store for document analysis.agent-teleport: Serialize agent state to cloud for cross-device migration.
Each skill is standalone. To give your agent the power to verify SQL:
# Install dependencies
pip install pymysql
# Run the skill (Agent simulates this)
python skills/verify-code/run.py --sql "CREATE TABLE test (id INT); INSERT INTO test VALUES (1);"Output:
{
"verified": true,
"infrastructure": "Powered by TiDB Cloud Zero"
}- Zero Config: No Docker, no cloud accounts required for the Agent.
- Ephemeral: Databases auto-recycle after 30 days.
- Standardized: All skills output strict JSON for easy parsing.