Ab Initio Log Recovery Agent (RAG)

This project was rebuilt as a focused agentic RAG assistant for Ab Initio graph failures.

You provide raw execution logs (like your g6t_policy_trans_src failure sample), and the agent:

1. Parses graph metadata + failed record fields.
2. Optionally retrieves the closest schema/process definition from a schema knowledge base.
3. Produces a recommended SQL UPDATE to reset process status and related fields so rerun is possible.

Why this helps when schema changes

Your schema/rules live in YAML knowledge docs (schemas/*.yaml). When table names, status columns, or rerun codes change, update YAML instead of code. The agent uses RAG retrieval over those docs to pick the right target behavior.

Installation

python -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt

Run with a log file (log-only mode)

python -m abinitio_sql_agent.cli \
  --log-file sample.log

Run with schema-assisted retrieval

python -m abinitio_sql_agent.cli \
  --schema schemas/example_schema.yaml \
  --log-file sample.log

Run with inline log text

python -m abinitio_sql_agent.cli \
  --log-data "Ab Initio Graph Execution Log ..."

Schema knowledge format

schemas:
  - name: g6t_policy_trans_src_prod
    table: G6T_POLICY_TRANS_STD
    process_status_column: process_status_code
    failed_value: FAILED
    rerun_ready_value: READY_FOR_RERUN
    id_columns: [trans_id, policy_id]
    mutable_columns: [process_status_code, retry_count, error_message, updated_by]
    notes: Main policy load schema

Output

The CLI prints:

• SQL update template
• bind parameters extracted from failed record fields
• retrieval reasoning + matched schema docs

In log-only mode (no --schema), the agent does not enforce fixed status checks from a schema doc. It lets the model infer a pragmatic update using table and failed-record fields extracted from the log.