8000 Design LLM provider abstraction layer · Issue #18 · adeebahmed/Libertas · GitHub
[go: up one dir, main page]

Skip to content

Design LLM provider abstraction layer #18

Open

Description

@adeebahmed

Design a provider-agnostic abstraction for LLM calls in the backend. Must support: swappable providers (Anthropic, OpenAI, Ollama), per-provider config (API keys, base URL, model name), and a unified response interface for the insights engine.

Activity

  1. added this to the Multi-Model milestone on Apr 11, 2026
  2. added
    aiAI/ML features and model work
    planningArchitecture and planning tasks
    on Apr 11, 2026
  3. adeebahmed commented on Apr 12, 2026

    @adeebahmed
    OwnerAuthor

    Planned for Phase 4: AI & Multi-Model Intelligence

    Architecture decision from research:

    • Provider abstraction pattern: Single interface with Claude, OpenAI, Ollama implementations
    • Critical constraint: Summarization/RAG layer required before sending to any LLM — large transaction histories will hit context limits without it
    • Recommended default: Rolling 90-day transaction window + pre-computed account summaries (net worth, allocation, recent spending patterns)
    • Rule engine does math; LLM explains patterns and answers narrative questions
    • All API keys stored locally, never logged or transmitted

    Phase 4 will implement this interface along with #19 (Claude), #20 (OpenAI), #21 (Ollama), #22 (model UI), #23 (context management).

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    aiAI/ML features and model workmvpRequired for MVP releaseplanningArchitecture and planning tasks

    Projects

    • Status
      Backlog

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions

    0