feat: backend for timmy

[ericfitz]

This is the server side work for Timmy

• add chat API for clients
  • api endpoint + schema
  • integrate with LLMs (configs + functionality)
  • integrate with vector store
• add vector store
  • integration with vector store database
  • add vectorization & indexing

[ericfitz]

Early Design Thoughts

[ericfitz]

Timmy: Purpose and Capabilities Summary

What it is: Timmy is a conversational AI assistant embedded in the threat modeling workflow. It's scoped to a single threat model — you talk to Timmy about a specific threat model, and it has deep context on that model's data: its assets, threats, diagrams, documents, repositories, and notes.

The NotebookLM analogy: Like Google's NotebookLM, Timmy is a "grounded" chat — it doesn't just answer from general knowledge, it reasons over the specific sources the user has loaded. Users control which sub-entities are included (via the timmy_enabled flag), so they can focus the conversation on relevant material.

Problems it solves:

1. Threat models are dense and hard to reason about holistically. A mature threat model has dozens of assets, threats, data flows, and supporting documents. Humans struggle to hold all of that in their heads simultaneously. Timmy can synthesize across the full model and surface connections, gaps, or inconsistencies a person might miss.

2. Security review is bottlenecked on expert availability. Not every team has a senior security reviewer on hand. Timmy acts as an always-available collaborator — it can't replace a human reviewer, but it can help teams self-serve on initial analysis, ask better questions, and arrive at a review better prepared.

3. Threat modeling artifacts are underutilized after creation. Teams build threat models and then rarely revisit them conversationally. Timmy makes the model queryable: "What are the highest-risk data flows?", "Which assets lack mitigations?", "Summarize the threats related to authentication."

4. Onboarding to an existing threat model is slow. A new team member or reviewer joining a threat model has to read through everything. Timmy can provide guided summaries and answer targeted questions, dramatically reducing ramp-up time.

How customers use it: A user navigates to a threat model's chat page, sees their sources (sub-entities) in a sidebar, toggles which ones to include, and has a conversation. They can ask Timmy to analyze threats, identify gaps, explain data flows, suggest mitigations, or just summarize what's there. Previous chat sessions are preserved and can be resumed.

Implementation approach: Provider-agnostic LLM abstraction using LangChainGo, with a vector store for grounding responses in threat model content.

See also: ericfitz/tmi-ux#293 for client-side design thoughts.