A set of tools for managing and using templated prompt flows across different data sources and models.
Fun. Privacy. Choice. Customisation. Specific domain and tribal knowledge.
- Extensible template libray from which flows can be created and customised
- Connectivity to atlasssian JIRA & Confluence (todo: s3, Office, Github)
- Document upload support for common docs (PDF, PPT, DOC, XLS, CSV)
- Prompt flow sessions. Create, duplicate, run reset, export JSON.
- todo: Publish and share flows for others to use via UI or API.
Data: Authorised, connected, indexed. Exploration: Templated processing of data with a selection of models. Output: Sharable. Publishable.
Definitions of common patterns that bring data sources, user input, LLM models and (todo: actions) Extensible format for customisation using YAML Actiosn and Anchors and a custom reference key words
An instanciation of a template. Has a series of steps each with an input, state and an putput. Create, manage and interact with flows via the UI (todo: and API)
Driven through the configuration, allow connection to data sources.
These are published to the GitHub container registry. Various intallation options.
A UI create and manage flows from templates.
Todo: An API to access and use flows.
Published in
Ready to use reference implementations
- todo: CDK scripts to deploy as AWS ECS
Custom configuration covers:
- Prompt Template Library (local, s3 or todo:github)
- State Storage Location (local or s3)
- Connectivity Secrets (atlassian, todo: Office365)
- Model Library - the flavours of LLMs that can be used
Specified at deploy time as a pointer to an AWS SSM param store path.