Wiki source code of LLM Abstraction Architecture (v2.6.33)
Last modified by Robert Schaub on 2026/02/08 08:31
Hide last authors
| author | version | line-number | content |
|---|---|---|---|
 |
1.1 | 1 | {{info}} |
| 2 | **Current Implementation (v2.6.33)** - Uses Vercel AI SDK for multi-provider abstraction. Provider selected via ##LLM_PROVIDER## environment variable. | ||
| 3 | {{/info}} | ||
| 4 | |||
| 5 | = LLM Abstraction Architecture = | ||
| 6 | |||
| 7 | {{mermaid}} | ||
| 8 | |||
| 9 | graph TB | ||
| 10 | subgraph Pipelines[Triple-Path Pipelines] | ||
| 11 | ORCH[Orchestrated orchestrated.ts] | ||
| 12 | CANON[Monolithic Canonical] | ||
| 13 | DYN[Monolithic Dynamic] | ||
| 14 | end | ||
| 15 | |||
| 16 | subgraph AISDK[Vercel AI SDK] | ||
| 17 | SDK[AI SDK Core generateText generateObject] | ||
| 18 | STREAM[Streaming Support streamText] | ||
| 19 | end | ||
| 20 | |||
| 21 | subgraph Providers[LLM Providers] | ||
| 22 | ANT[Anthropic Claude 3.5 Sonnet DEFAULT] | ||
| 23 | OAI[OpenAI GPT-4o] | ||
| 24 | GOO[Google Gemini 1.5] | ||
| 25 | MIS[Mistral Large] | ||
| 26 | end | ||
| 27 | |||
| 28 | subgraph Config[Configuration] | ||
| 29 | ENV[Environment Variables LLM_PROVIDER FH_DETERMINISTIC] | ||
| 30 | end | ||
| 31 | |||
| 32 | ORCH --> SDK | ||
| 33 | CANON --> SDK | ||
| 34 | DYN --> SDK | ||
| 35 | SDK --> ANT | ||
| 36 | SDK --> OAI | ||
| 37 | SDK --> GOO | ||
| 38 | SDK --> MIS | ||
| 39 | ENV --> SDK | ||
| 40 | |||
| 41 | {{/mermaid}} | ||
| 42 | |||
| 43 | == Current Implementation == | ||
| 44 | |||
| 45 | |= Feature |= Status |= Notes | ||
| 46 | | **Multi-provider support** | Implemented | Anthropic, OpenAI, Google, Mistral | ||
| 47 | | **Provider selection** | Implemented | Via ##LLM_PROVIDER## env var | ||
| 48 | | **Deterministic mode** | Implemented | ##FH_DETERMINISTIC=true## sets temperature 0 | ||
| 49 | | **Automatic failover** | Not implemented | Manual provider switch only | ||
| 50 | | **Per-stage provider** | Not implemented | Single provider for all stages | ||
| 51 | |||
| 52 | == Environment Variables == | ||
| 53 | |||
| 54 | |= Variable |= Default |= Options | ||
| 55 | | ##LLM_PROVIDER## | anthropic | anthropic, openai, google, mistral | ||
| 56 | | ##FH_DETERMINISTIC## | true | true = temperature 0, false = default | ||
| 57 | |||
| 58 | == Provider Details == | ||
| 59 | |||
| 60 | |= Provider |= Model |= Use Case | ||
| 61 | | **Anthropic** | Claude 3.5 Sonnet | Default, best reasoning | ||
| 62 | | **OpenAI** | GPT-4o | Alternative, fast | ||
| 63 | | **Google** | Gemini 1.5 Pro | Alternative, long context | ||
| 64 | | **Mistral** | Large | Alternative, EU data residency | ||
| 65 | |||
| 66 | == Future Enhancements == | ||
| 67 | |||
| 68 | * **Automatic failover**: Chain providers for resilience | ||
| 69 | * **Per-stage optimization**: Different providers per pipeline stage | ||
| 70 | * **Cost tracking**: Monitor and optimize LLM costs | ||
| 71 | * **Local models**: Ollama/vLLM for on-premises deployment |