Conversation Engine

Keep a conversation source and its evolving transformations consistent, traceable, and recoverable in a local vault.

Try it

Python 3.12+. Run from this checkout:

python -m pip install -e .
python -m examples.offline_demo

Input: A source message, an explicitly configured Workshop mapping, and three synthetic worker replies.

Result: All six stages acquire current results; level 9 resolves to “Review the fixture.” The demo creates and removes its own disposable vault.

Read the complete synthetic worker exchange: input envelope, materialized input filenames, exact returned files and resolved outputs. The guided path connects that run to the contracts.

How it works

Each workspace captures the contract and definitions it must satisfy. The validator promotes results only when that captured work remains relevant. Current outputs can reference source or other results instead of copying text.

Source: conversation_engine/service.py, conversation_engine/cli.py, tests/test_workspaces_and_contracts.py.

Use it for your work

For your own vault, run conversation-engine --root YOUR_VAULT init, inspect verify and status, then follow the contracts and workspace code before connecting a worker. Source data and runtime stay local.

Scope

The engine does not supply a semantic model. The demo supplies labeled fixture replies; real operation needs a feeder, an approved mapping definition and a worker that returns the documented files. New vault intake begins paused. Use a short runtime root on Windows.

Owned code is available under the MIT license.

Present edition

This is a local transformation-engine snapshot with an executable, synthetic end-to-end workflow. It preserves the source, versioned definitions and captured contracts; it does not provide the semantic worker. A Public fixture decision demonstrates how the engine accepts a correctly formed response, not that the message was independently evaluated for disclosure.

Start with the worked flow, then stage contracts and operations. The existing architecture is the detailed source map. Origin identifies the snapshot; the expanded fixture trace is public continuation documentation, not recovered historical execution.

ChatGPT Export Archive prepares optional intake envelopes. Conversation Archive for Discord is a separate consumer. Feeders, workers and consumers remain outside this engine. Real-worker evaluation, deployment and migration each require their own evidence; they are not implied by the offline run.