SimpleFunctions Portfolio Autopilot Setup

When to use

• First-time autopilot enablement
• Reconfiguring an existing autopilot (new strategy, tighter risk gates)
• Setting up after a halt to re-engage with adjusted parameters

Safety posture

Default is human-in-the-loop. Autopilot proposes orders, human approves through the dashboard or sf intent commands. Closed-loop (autonomous execution) requires explicit opt-in per thesis AND a global toggle.

Before enabling closed-loop, confirm:

• Risk gates are configured (max exposure, max concentration, max daily loss)
• Drawdown auto-halt is on
• BYOK credentials are encrypted at rest

Prerequisites

npm install -g @spfunctions/cli
sf login

You need:

• A Kalshi account with API access enabled
• (Optional) Polymarket wallet for cross-venue
• An OpenRouter API key for the autopilot's LLM

Instructions

Step 1: Connect Kalshi credentials (BYOK)

sf portfolio enable --venue kalshi

This launches an interactive flow:

1. Prompts for Kalshi API key + secret
2. Encrypts client-side before upload
3. Stores in your account vault (cloud-encrypted)
4. Verifies by fetching account balance

Confirm with:

sf portfolio cloud status --json | jq .credentials

Should show kalshi: { connected: true, lastVerified: "..." }.

Step 2: Set OpenRouter key

sf portfolio cloud config --openrouter-key "sk-or-..."

This is also encrypted at rest. Used by autopilot for evaluation LLM calls.

Step 3: Set risk gates

sf portfolio cloud config \
  --max-exposure-usd 5000 \
  --max-position-pct 10 \
  --max-daily-loss-usd 200 \
  --drawdown-auto-halt-pct 15 \
  --json

These are pre-trade — orders that would violate are rejected before submission to the exchange. See concepts/risk-gates for the full list.

Step 4: Create a view (investment universe)

sf portfolio view create \
  --name "macro-events" \
  --filter '{
    "categories": ["macro", "geopolitics"],
    "venue": "kalshi",
    "min_volume_24h": 1000,
    "max_days_to_event": 60,
    "min_iy": 100
  }' \
  --json

A view is the set of markets the autopilot considers. Without a view, it considers everything (noisy).

Common view patterns:

• macro-events: macro + geo, high volume, 30-60 day window
• cross-venue-arb: high-confidence pairs only
• term-structure-binary: same event, multiple deadlines (see sf-research skill for term structure analysis)

Step 5: Attach a strategy

sf portfolio strategy create \
  --view macro-events \
  --name "momentum" \
  --rules '{
    "entry": "vr > 2 AND price between 20 and 80",
    "size_pct": 5,
    "exit": "vr < 0.5 OR price hits target",
    "max_holds": 5
  }' \
  --json

Strategy = entry/exit/sizing rules applied within the view. Multiple strategies can target the same view.

Step 6: Set cadence + budget

sf portfolio cloud config \
  --tick-cadence-hours 12 \
  --monthly-budget-usd 50 \
  --json

tick-cadence-hours defaults to 12 — that's the intended cadence (per feedback_portfolio_tick_cadence memory). Don't drop below 4h without specific reason (LLM cost).

Step 7: Start in observation mode

Run the first tick manually to verify everything works:

sf portfolio cloud tick --dry-run --json

Returns the handoff note + proposed actions WITHOUT placing orders. Review the proposal. If sensible, proceed.

Step 8: Enable scheduled ticks (still human-in-the-loop)

sf portfolio cloud start --json

The cloud cron now fires every tick-cadence-hours. Each tick:

1. Pulls latest market state for the view
2. LLM proposes actions (entries, exits)
3. Risk gates filter proposals
4. Approved proposals become sf intent objects with autoExecute: false
5. You review in dashboard or via sf intent list
6. Approve with sf intent approve <id> → order goes to exchange

Step 9: (Optional) Enable closed-loop per thesis

If you want autopilot to execute without human approval:

sf thesis fork <thesis-id> --closed-loop-entry true --closed-loop-exit true

Closed-loop is per-thesis, not global. The thesis must still be within the autopilot's active view + strategy.

Strongly recommended: leave closed-loop off until you've watched 1-2 weeks of human-approved ticks.

Verification

sf portfolio cloud status --json | jq '{status, lastTickAt, nextTickAt, monthlySpend, riskGates}'
sf intent list --json --limit 10  # pending intents waiting for approval

Errors and recovery

• Credentials encryption fails: ensure local CLI version is up to date. npm i -g @spfunctions/cli@latest.
• Risk gate rejects every proposal: gates too tight for view. Loosen one at a time, watch behavior.
• `status: halted` after first tick: check haltedReason. Often: budget exhausted (raise monthly-budget-usd) or no markets matched view (broaden filter).

Related skills

• sf-portfolio-review — audit what autopilot has done
• sf-thesis-monitor — per-thesis monitoring config
• sf-pre-trade-check — adversarial check for high-value entries (use before approving sf intent)

Reference docs

• <https://docs.simplefunctions.dev/guides/portfolio-autopilot>
• <https://docs.simplefunctions.dev/concepts/risk-gates>
• <https://docs.simplefunctions.dev/cli/portfolio>
• <https://docs.simplefunctions.dev/build/trade-intents>