Skip to content

CHECKPOINTS.md

Latest commit

 

History

History
73 lines (49 loc) · 3.11 KB

CHECKPOINTS.md

File metadata and controls

73 lines (49 loc) · 3.11 KB
Raw

OnyxAI Implementation Checkpoints

Checkpoint 1: Foundation

  • Replace the temporary Express prototype with the spec architecture.
  • Scaffold Expo Router, route tree, shared constants/types/store, Worker folders, and Supabase folders.
  • Create env placeholders for mobile and Worker secrets.

Validation prompt: Confirm the project structure, scripts, and env placeholders look right before wiring real data.

Checkpoint 2: Core App Shell

  • Build auth, chat, history, memory, credits, and settings screens.
  • Add reusable UI components, model selector, message list, and input bar.
  • Keep everything runnable in preview mode without real credentials.

Validation prompt: Open the Expo app and confirm screen structure, navigation, and visual direction before backend wiring.

Checkpoint 3: Client Data Wiring

  • Add Supabase client integration and auth session handling.
  • Connect Zustand state, conversation hooks, memory hooks, and credit hooks.
  • Replace preview-only data with real reads where credentials exist.

Validation prompt: Test sign-in, conversation loading, and balance fetching before enabling model streaming.

Checkpoint 4: Worker + Billing + Memory

  • Implement Worker auth verification, /chat, /memory/extract, and /payments/webhook.
  • Connect OpenRouter streaming, usage recording, and credit deduction.
  • Add memory extraction and retrieval flow.

Validation prompt: Verify worker auth, streaming responses, and credit deduction against test accounts before shipping.

Checkpoint 5: Hardening

  • Finish DB migrations, error states, polish, and deployment config.
  • Add tests and a final production checklist.

Validation prompt: Run through staging on device before enabling real top-ups or wider testing.

Wave 1: Context-Window Management (Complete)

Date: 2026-05-13

Changes

  1. DB Schema (supabase/migrations/0004_wave1_context_window.sql)

    • Added conversation_summaries table with GIN index on key_facts.
    • Added pgvector extension + match_messages RPC for semantic retrieval.
    • Added embedding vector column to messages.

  2. Worker (worker/src/)

    • New POST /chat/summarize endpoint (handleSummarize) — runs summarization on the 2nd-last 10-message chunk using the cheapest available model.
    • Updated POST /chat (handleChat) — prepends summary context as a system message when conversationId is provided.
    • Added fetchConversationSummaries, fetchMessagesRange, fetchMessageCount, insertConversationSummary, matchMessages helpers in worker/src/supabase.ts.

  3. Mobile (hooks/useChat.ts, lib/openrouter.ts, lib/tokens.ts)

    • useChat trims request history to last 8 messages before sending to worker.
    • Auto-triggers summarizeConversation() every 10 messages (fire-and-forget, errors swallowed).
    • estimateTokens() updated to use meta-llama/llama-4-scout:free (128k context) as default.
    • Added summarizeConversation() helper in lib/openrouter.ts.

Validation prompt:

Create a conversation with >10 messages and verify summaries appear in the DB + context is injected into new /chat requests.