Quickstart¶

Outcome: Send a chat completion and see an L0 cache hit on the second identical request.

Prerequisites¶

daari installed (Install)
Daemon running: daari serve (or docker compose up)
For live local models: Ollama with llama3.2:3b (Compose pulls this for you)

Steps¶

1. Health check¶

curl -fsS http://127.0.0.1:11435/health

2. First completion¶

curl -s http://127.0.0.1:11435/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "X-Daari-Meta: true" \
  -d '{"model":"daari","messages":[{"role":"user","content":"Say hi in one word"}]}'

Expect daari_meta.tier of L3–L5 (or L0/L1 if already cached).

3. Repeat — cache hit¶

Run the same curl again. Expect "tier": "L0" and "cache_hit": true in daari_meta.

4. Stats¶

daari stats
# or
curl -s http://127.0.0.1:11435/v1/daari/stats | python -m json.tool

One-shot demo¶

./scripts/demo.sh

Troubleshoot¶

Symptom	Fix
Connection refused	Start `daari serve` or Compose
`/ready` fails	Ollama not up or model not pulled
Always L3, never L0	Identical messages required; check `X-Daari-No-Cache` is unset
401 Unauthorized	`server.api_key` set — send `Authorization: Bearer <key>`

Next¶

→ First client · Routing tiers