mirror of
https://github.com/diegosouzapw/OmniRoute.git
synced 2026-08-08 00:02:20 +03:00
f44ec7e1f2
T-30 — ADRs:
- 6 ADRs: SQLite, Fallback Strategy, OAuth, JS+JSDoc, Single-Tenant, Translator Registry
T-33 — JSDoc Coverage:
- Full JSDoc on all new modules (100% exported functions documented)
T-35 — Accessibility:
- a11yAudit.js: lightweight WCAG AA checker (aria-label, dialog role, alt text, labels)
T-38 — Password Reset CLI:
- bin/reset-password.mjs: interactive CLI tool for admin password reset
T-39 — Playwright Specs:
- tests/e2e/responsiveSpecs.mjs: viewports (375/768/1280), 4 pages, test matrix
T-42 — Eval Framework:
- evalRunner.js: 4 strategies (exact, contains, regex, custom) + golden set (10 cases)
T-43 — Compliance:
- audit_log table, noLog opt-out per API key, LOG_RETENTION_DAYS cleanup
TASKS.md: 46/46 Concluído ✅
Tests: 144/144 pass (119 existing + 25 new)
1.6 KiB
1.6 KiB
ADR-006: Translator Registry Pattern
Date: 2025-12-01
Status: Accepted
Deciders: @diegosouzapw
Context
OmniRoute translates requests between different LLM API formats (OpenAI ↔ Anthropic ↔ Google ↔ etc.). Each provider has a unique request/response schema. The translator must:
- Convert incoming requests to the target provider's format
- Convert streaming responses back to the client's expected format
- Handle provider-specific features (tool calls, vision, system prompts)
Decision
Use a registry pattern for translators:
- Each provider pair has a translator module in
src/sse/translators/ - Translators are registered by
(sourceFormat, targetFormat)key - The
translateRequest()function auto-detects source format and applies the appropriate translator - Translators handle both request translation and response stream mapping
Key translators:
openai → anthropic(and reverse)openai → google(and reverse)anthropic → google(and reverse)- Identity translators for same-format routing
Consequences
Positive
- Adding a new provider requires only a new translator module
- Each translator is independently testable
- Auto-detection reduces configuration burden on users
- Supports chained translation (A → B → C) if needed
Negative
- O(n²) translator combinations as providers grow (mitigated by identity translators)
- Some edge cases in format conversion (e.g., tool call schemas differ significantly)
Neutral
- The Translator Playground UI provides visual testing of translation chains
- Performance overhead is minimal (JSON transformation, no network calls)