-
Notifications
You must be signed in to change notification settings - Fork 0
Evaluate LiteLLM: integration prototype, limitations, alternatives #4
Copy link
Copy link
Closed
Labels
prio:criticalBlocks other work, must do firstBlocks other work, must do firstscope:medium1-3 days of work1-3 days of workspec:providersDESIGN_SPEC Section 9 - Model Provider LayerDESIGN_SPEC Section 9 - Model Provider Layertype:researchEvaluate options, make tech decisionsEvaluate options, make tech decisions
Milestone
Description
Context
LiteLLM is the primary candidate for provider abstraction (spec 9.3). Before committing to it, we need a thorough evaluation to understand its capabilities, limitations, and whether it meets our requirements across all target providers.
Acceptance Criteria
- Evaluate all target providers supported: Anthropic, OpenRouter, Ollama
- Verify cost tracking reliability across providers
- Test retry and fallback behavior
- Test tool calling consistency across providers
- Measure overhead: latency impact, memory usage
- Verify streaming support quality
- Test down-provider handling and error propagation
- Evaluate alternatives: direct SDKs, aisuite, custom abstraction
- Working prototype with 2+ providers demonstrating core flows
- Cost tracking verified against known pricing
- Retry mechanism tested with simulated failures
- Tool calling tested across at least 2 providers
- Decision documented with rationale
Dependencies
- Depends on Design unified provider interface (base classes, protocols) #3 (unified provider interface design)
Design Spec Reference
Section 9.3 — LiteLLM Integration
Reactions are currently unavailable
Metadata
Metadata
Assignees
Labels
prio:criticalBlocks other work, must do firstBlocks other work, must do firstscope:medium1-3 days of work1-3 days of workspec:providersDESIGN_SPEC Section 9 - Model Provider LayerDESIGN_SPEC Section 9 - Model Provider Layertype:researchEvaluate options, make tech decisionsEvaluate options, make tech decisions