One assistant. Thirty-six engines underneath.

Deep-dive: a request that needed the failover

The Coder example above, full trace of what happens when the primary engine is rate-limited and the system hands off to the next.

Full walkthrough with Mermaid diagrams: How Failover Works · Architecture Diagrams

v2 request lifecycle

A single message now passes through the intent auto-router, optionally hands off to the agent runner, drives the TTS or STT cascade if voice is involved, hits the live-data tools where intent calls for them, and returns three smart follow-up suggestions alongside the answer.

-- Schema · supabase/migrations/001_sarmalink_ai.sql
create table ai_user_memories (
  id          uuid primary key default gen_random_uuid(),
  user_id     uuid not null references auth.users(id) on delete cascade,
  fact        text not null,
  embedding   vector(384),
  created_at  timestamptz default now()
);

alter table ai_user_memories enable row level security;
create policy "own_memories" on ai_user_memories
  for all using (auth.uid() = user_id);

-- Retrieval at request time
select fact from ai_user_memories
  where user_id = auth.uid()
  order by created_at desc
  limit 30;