How to use Jarvis
Your offloaded bulk-thinking brain
1 What this is
Jarvis runs 1000s of LLM jobs on a server (DeepSeek) so your laptop never freezes. You are the orchestrator: dispatch bulk work here, keep the hard reasoning for yourself. The server never calls Claude.
2 Connect
Paste your X-API-Key top-right β Connect. It's stored only in this browser. Auth is required for every action (no key β 401).
3 When to use it
Reach for the brain on high-volume or parallel work β crawling, summarizing many items, classifying, drafting variants, boilerplate. Keep nuanced judgment & final synthesis in your own session.
4 Personas
A persona is a reusable prompt template. Its payload fills the {{placeholders}} it expects:
β’ researcher β {"topic": "..."}
β’ crawler β {"content": "..."}
β’ frontend_builder β {"request": "..."} (returns HTML β use the Render tab)
5 Submit & read
Submit queues one job; Γ10 fires a batch to stress-test parallelism. Jobs flow queued β running β succeeded. Click any row to read the output; HTML results get a live Render tab.
6 Params (optional)
Override per-job: {"max_tokens": 800, "temperature": 0.2}. Leave {} to use the persona's defaults.
7 Cost
The spend counter (top bar) is live total cost across succeeded jobs. Full breakdown at GET /usage. DeepSeek is ~20Γ cheaper than premium APIs β bulk work costs fractions of a cent.
From code? Same API over HTTPS with the X-API-Key header. POST /jobs or /jobs/batch, then GET /jobs/{id}. See USE_THIS_IN_SESSIONS.md to paste into a Claude session.
+ Add a persona
# personas/my_persona.yaml
name: my_persona
template: |
Do X with {{thing}}.
default_params: {max_tokens: 600, temperature: 0.4}
Drop it on the server, restart the gateway, and it appears in the dropdown.