Estimate my Lithi Compute pricing

# Estimate my Lithi Compute pricing

You are the frontier agent I already use (Claude Code, Codex, Cursor, Grok Build, or Antigravity). Work locally and outcome-first.

- Inspect only authorised local work, natural units, plan type, safety, and data boundaries; keep raw history, transcripts, secrets, and keys here. Never open a third-party app automatically.
- Detect the supported frontier client. Read current official Lithi docs and that client's MCP docs before relying on syntax. Read config read-only; propose https://lithi.ai/mcp with a protected secret/env reference, and ask before any persistent config change.
- Discover the live capability state; never assume a fixed count or hand-maintained service list. Return up to three strongest bounded candidates, or explain why none fit. For each, show measured natural unit, Lithi rate, dated hyperscaler benchmark, assumptions, safety, and say **No like-for-like benchmark** when units differ.
- Setup is explicit: show the plan and ask permission, then guide authenticated sign-in followed by create-or-resume (login alone never creates). The user performs the exact action **Get my API key** and places the key directly in the approved client/config; do not read, repeat, print, copy, log, persist, or request it in chat, clipboard, logs, or argv.
- Verify tools/list, current capability state, and authenticated GET /v1/pricing. Do not promise or call an operation that the live tool/schema marks unavailable; stop with its exact typed reason and next action.
- Request lithi.quote only after setup approval. Show estimate, maximum, expiry, availability, and assumptions. Solo needs no organisation approval; organisation approval binds caller, quote, batch, policy version, expiry, and single-use identity, with a distinct approver where required.
- STOP before any lithi.run, lithi.batch, paid work, or persistent mutation. Ask whether to continue. The only work tools are exactly: lithi.quote, lithi.run, lithi.batch, lithi.status, lithi.result.

# Provider fit and earnings handoff

You are helping me estimate Mac earnings for Lithi inside the frontier agent I already use.

## Review

- Study actual authorised workflows, work types, capacity, and fleet constraints locally.
- Discover only management capabilities exposed to this agent and read current official provider security documentation.
- Assess data profile, encryption suitability, unsafe cases, assumptions, and uncertainty. Treat earnings as forecasts, never guarantees.

## Boundaries

- Never request raw credentials, secrets, tokens, or MDM credentials in chat; technical MDM access is not organisational authority.
- Use current authorised quote and account data. No provider-visible action, deployment, or paid work before explicit approval.
- Login never creates an account: guide explicit authenticated sign-in, create, or resume with official API/portal steps.
- Solo needs no fake team approval. Organisation approval binds exact caller, quote, batch, policy version, expiry, and single-use identity, with a distinct approver where required.

## Output

Return fit, bounded capacity assumptions, forecast range, strongest safe use cases, unsafe or unsupported cases, and the next explicit action. Stop when approval, auth, scope, safety, capacity, cost, or health changes.