PacificaPilot is a local-first agent for Pacifica perpetual futures. It combines multi-provider AI reasoning with live technical analysis, social sentiment, and persistent memory via Supermemory — so your agent remembers every trade, every preference, and every pattern across sessions.
01 — Install
PacificaPilot ships as a Python package (3.11+). It runs entirely on your machine — your keys never leave the box. Run pacifica init once to set keys, config, and optional memory.
pipx install pacificapilot02 — Capabilities
A focused set of primitives — not a bloated platform. Each capability composes into the agent's tool loop.
Supermemory stores every trade, pattern, and preference. Said once, remembered forever — across sessions and models.
Claude Code-style multi-turn reasoning. The AI gathers data, calls tools, reasons step-by-step, then responds.
24/7 market monitoring with AI-driven trading decisions. Every decision written to memory automatically.
Place orders, close positions, check balances, run TA, detect market regime, pull performance metrics.
RSI 5m/1h, MACD, Bollinger Bands, funding rates, volume signals, and regime detection wired into every prompt.
Anthropic, OpenAI, Google Gemini, or OpenRouter. Swap providers at runtime with /apikey. No lock-in.
Every order requires explicit yes/no. Dry-run by default. Testnet first. Nothing hits the exchange without you.
Your keys stay on your machine. With local Supermemory mode, zero data ever leaves the box.
Fixed 3-panel layout, slash autocomplete, live status sidebar, trade confirmation modals — not a debug REPL.
03 — The Supermemory Advantage
Every other trading agent starts from zero every session. Open the CLI, and the AI has no idea what it traded yesterday, what patterns it noticed, or what you told it. This causes hallucination, repetitive questions, and zero learning.
PacificaPilot integrates Supermemory as a persistent, searchable memory layer. Every trade decision, pattern, and preference is stored under container tags and injected into the AI's system prompt on every session.
Hosted by Supermemory — no local process needed.
Runs on localhost:6767. Zero data leaves.
"LONG BTC at $50k, confidence 75%, RSI was 35"
"Funding rate spike on ETH: 0.003 at 14:30"
"User never wants to trade BONK"
"Daily: 3 trades, +$12.50 USDC, 67% win rate"
"Binance fallback activated for BTC at 14:30"
04 — Architecture
A Loop Agent runs autonomously on a timer, fetching live data and making AI-driven decisions. A Chat Agent handles your prompts with a multi-turn tool loop. Both share the same trading core and write to Supermemory.
Multi-turn tool loop. Gathers data, reasons step-by-step, then responds. Up to 8 iterations per turn.
Autonomous 24/7 monitoring. Fetches data, decides trades, writes every decision and pattern to memory.
Supermemory — 5 container tags, semantic recall, profile injection. Cloud or fully local.
User → "How are BTC markets, should I trade?"
↓
[1] AI decides to call get_market_price("BTC")
[2] Execute → { price, RSI, MACD, Bollinger, volume, funding }
[3] Feed result back to AI (not user)
[4] AI calls get_market_regime("BTC")
[5] Execute → { regime: "weak uptrend", guidance: ... }
[6] Feed result back to AI
[7] AI has full context — produces final response
↓
User ← "BTC is in a weak uptrend. RSI neutral, MACD bullish.
Volume confirms. I'd hold for now."05 — Natural Language
No DSL, no rigid syntax. Slash commands when you want structure — plain English for everything else.
06 — Tools
The AI decides which of these to call — you never have to remember function signatures. Trade actions always require confirmation.
| Tool | What it does |
|---|---|
| place_order | Open LONG/SHORT positions (with confirmation) |
| close_position | Close a specific position |
| close_all_positions | Flatten the entire book (batch confirmation) |
| get_positions | View open positions with unrealized PnL |
| get_account_balance | Check USDC balance, equity, available capital |
| get_market_price | Full snapshot: price, RSI, MACD, Bollinger, volume, funding, regime |
| get_trade_history | Past trades with PnL, duration, entry/exit |
| get_performance_metrics | Win rate, Sharpe, profit factor, drawdown |
| get_market_regime | Detect trending/ranging/volatile with trading guidance |
07 — Comparison
| Feature | PacificaPilot | Other bots |
|---|---|---|
| Persistent memory | Cross-session via Supermemory | None — starts from zero |
| AI providers | BYOK: Anthropic, OpenAI, Google, OpenRouter | Usually one vendor |
| UI | Textual TUI — 3-panel, live sidebar | Basic CLI prints or web dashboard |
| Reasoning | Multi-turn agentic tool loop | One-shot ask & respond |
| Data locality | 100% local, non-custodial | Often a hosted backend |
| Technical analysis | Live RSI, MACD, Bollinger, funding, regime | Basic or none |
| Install | pip install pacificapilot | Docker / Kubernetes setup |
| Cost | Free + your API keys | Monthly SaaS fees |
08 — Security
Keys stored on your machine, nowhere else.
Your API keys go directly to your provider — no proxy.
Every trade action requires explicit yes/no.
All trades simulated by default until you flip the switch.
New setups start on testnet — no real money at risk.
Supermemory is additive. Trading works without it.
09 — Setup
pipx install pacificapilotpacifica init
# → Pacifica keys → AI provider → config → Supermemorypacifica config set anthropic.api_key sk-ant-...pacifica start10 — Reference
| Command | Description |
|---|---|
| /start | Boot the autonomous Loop Agent |
| /stop | Stop the Loop Agent |
| /pause | Soft-pause the loop |
| /resume | Resume the loop |
| /config | View/edit trading parameters |
| /apikey | Manage AI provider and Supermemory keys |
| /mode | Switch testnet/mainnet |
| /status | View system status |
| /help | Show all commands |