URGENT GAZA HUMANITARIAN EMERGENCY: Support vital civilian relief efforts directly through trusted global organizations.
♡ ISLAMIC RELIEF 🌐 UNRWA
ⓘ
SOURCE AVAILABLE · V2.4.0 · MULTI-AGENT ARCHITECTURE

CairnIQSelf-Hosted AI Portfolio Intelligence

A self-hosted AI financial analyst for private stock research and portfolio analysis. 6 AI agents, 125+ financial tools, real-time stock market data — all running locally. Your portfolio, your keys, your data.

Deploy Locally Explore Agent Architecture
Chat
CAIRNIQ // CONSOLE

6

AI Agents

125+

Financial Tools

0

Cloud Persistence

∞

Local Control

Multi-Agent System

Six Agents, One Unified Brain

Each agent is an expert in its domain. The Supervisor routes your queries to the right specialist, and the Deep Reasoning engine orchestrates multi-step analysis plans.

🧠

Deep Reasoning Engine

Breaks complex financial queries into multi-step execution plans. Selects from 125+ tools, runs them in parallel, and synthesizes institutional-grade reports. When a tool fails, a hand-curated equivalent stands in — and says so.

📊

Market Analyst

Real-time market intelligence. Scans for golden crosses, oversold conditions, volume spikes, and sector rotation signals across your entire universe.

⚠️

Risk Manager

Monitors concentration risk, drawdown limits, and macro regime shifts. Runs Monte Carlo simulations and position-sizing calculations.

View Full Agent Specs & Collaboration Diagram →
Command Center

Talk to Your Portfolio Like It Understands You

Natural language interface with context-aware intelligence panels. Ask complex questions and watch the AI break them into multi-step plans — live.

  • Quick Action bar: Today's Priority, Analyze, Scan, Dip Plan, Health, News, Risk
  • Live reasoning trace — watch the plan form, tool by tool
  • Real-time Portfolio vs. S&P 500 benchmark overlay
  • Daily Market Pulse with regime detection and a ranked catalyst feed
CairnIQ chat console with Quick Actions, live reasoning trace, and the Today's Priority, Market Pulse and news panels
Holdings

Two Ways In, One Source of Truth

Sync a broker, or keep your keys entirely to yourself and upload a CSV. Either way the editor is a live grid — last price, day move, computed return and total value, sortable, with a live quote always beating a pinned CSV price.

  • Questrade and Alpaca sync, or a CSV template from any broker
  • Overlapping positions merge automatically — live sync wins over manual rows
  • Multi-currency by design: totals in your base currency, trades quoted in the currency actually traded
  • ETFs and funds decompose into their real sector exposure, not "Unclassified"
CairnIQ portfolio editor grid showing symbols, units, entry price, last price, total value and return per position
Grounded Advice

Every Answer Is Audited Before You See It

An LLM that sounds certain is easy. The hard part is proving the numbers came from somewhere. Every advice-generating turn passes a review gate that is deliberately split between deterministic code and a challenge model — because the two fail differently.

⚖️

Risk Judge

A final challenge pass that hunts unsourced numbers, claims about holdings you don't own, and invented history. Every verdict is written to a per-profile audit trail.

📐

Deterministic Pre-Check

Proposed trades are extracted and checked numerically against your stated caps — position, sector, dollar-at-risk — before the judge ever reads the draft.

🔎

Turn Provenance

Every turn records what its evidence was worth: sources live, unavailable, stale, or unstamped. Thin evidence caps the answer's confidence instead of shipping like a fully-sourced one.

🧪

Regression Harness

A golden-set corpus of engineered drafts run against the real judge, so a model, provider, or prompt change can't quietly weaken the safety layer.

No house defaults, anywhere. The "2% risk rule" and the "10% concentration cap" that most tools ship with were removed, not documented — they were in nobody's profile, and the system was quoting them back as if you had set them. Limits come only from what you state. Unstated means unconstrained, and a rule you never wrote is never attributed to you.
Speaks First

The Advisor That Calls You

Between conversations, zero-LLM monitors watch the tape and your book, and anything worth saying lands in one deduplicated inbox — with a desktop notification for anything above informational.

  • Watch conditions: trigger levels the advisor committed to are stored and re-checked every 30 minutes
  • Intraday sentinel: fires on a market state change — never on a level that has stood all session
  • Event radar: earnings, ex-dividend and FOMC dates on names you hold — T-3 and T-1 only
  • Armed deployment ladder: the drawdown rungs you wrote while calm fire at the levels you named
  • Freshness gates: an alert refuses to fire on a quote that can't be proven recent
CairnIQ Proactive Intelligence inbox listing triggered alerts and active watch triggers
Accountability

It Keeps Score On Itself

An engine that is correct, running, and invisible is indistinguishable from one that is dead. These four surfaces exist purely to be read.

📒

Advisor Ledger

Every past call scored against what actually happened — 2-week, 1-month and 3-month returns vs SPY, plus how accurate its own stated confidence turned out to be.

🗞️

Weekly One-Pager

Assembled Sunday evening, waiting Monday. Every section always renders — including the ones whose honest content is "nothing this week."

🎛️

Readiness Switchboard

Which inputs only you can state are still blank — and exactly which shipped feature is switched off until you fill them. It never suggests a value.

💓

Engine Health

Whether each background monitor actually ran, reported as the count that proves the whole chain worked rather than a rare event that may legitimately be zero.

Advisor Ledger

Its Own Track Record, Kept Honestly

Every call is logged with its stated confidence and horizon, then scored against what the market actually did — and the arithmetic is defended as carefully as the advice.

  • Returns at 2 weeks, 1 month and 3 months, measured against SPY
  • Stated-confidence accuracy — did "high conviction" actually mean anything?
  • Statistics count distinct calls, never ledger rows, so a restated call can't inflate the sample
  • A call reversed before its horizon is graded at supersession, not quietly dropped
CairnIQ Advisor Ledger console displaying performance horizons, stated confidence accuracy, and logged calls table
Weekly Review

Weekly One-Pager & Engine Telemetry

Assembled Sunday evening, waiting Monday. Every section always renders — including the ones whose honest content is "nothing this week."

  • Wealth goal progress, market state briefings, and advice scoring in one view
  • Weekly summary of what the advisor said, verdict scores, and active holdings
  • Real-time background engine status — tracks consecutive runs, sentinel gates, and watch conditions
  • Nothing is inferred or filled with defaults: blank sections state clearly why they are unpopulated
CairnIQ Weekly Review console displaying wealth goal status, market state, weekly verdicts, and background engine telemetry
Holdings Monitor

Fund Flows, Event Radar & Asset Location

Continuous monitoring across fund creations/redemptions, earnings calendars, and macro event countdowns for names in your book.

  • Fund Flows: week-over-week creations and redemptions for funds you hold
  • Event Radar: real-time countdown to FOMC rate decisions and earnings reports
  • Asset Location: tax-efficient placement verification across taxable and tax-deferred accounts
  • Explicit data freshness — flags names with missing or unreturned provider dates
CairnIQ Monitor console showing fund flow tracking, FOMC & earnings event radar, and asset location rules
Silence is never back-filled. Empty sections used to vanish, and a reader expecting a full page filled the gap with assumptions — so a report once described history that never happened. Now an empty ledger says it is empty, a missing figure says why it's missing, and "nothing to report" is printed as a result rather than hidden as an absence.
Tool Arsenal

125+ Specialized Stock Market & Portfolio Tools

Every tool the AI uses is a real, deterministic function — not a hallucinated answer. From RSI calculations and stock screening to Monte Carlo simulations and portfolio analysis.

📡

Opportunity Scanner

Accumulation-first funnel: Dynamic Universe → Event & Flow Radar → Deep Fundamentals → Conviction Scoring

🔮

Monte Carlo Simulation

10,000-path probability engine for portfolio projections

📉

Tax-Loss Harvesting

Surfaces loss-harvesting candidates with a recovery-conviction check and wash-sale guidance

🏦

FRED & Bank of Canada Macro

US macro from the Fed, and keyless first-class CAD data straight from the institution that sets the rate — policy rate, CORRA, core CPI, posted bank rates

🕵️

Insider Filings, Both Venues

SEC Form 4 with true transaction coding, plus Canadian SEDI filings read in their own vocabulary — an issuer buyback is not an insider buy

🎙️

Earnings Call Tone

Loughran-McDonald scoring of management language, reported as the quarter-over-quarter change — a team that always hedges is not a sell

⚙️

Constrained Optimizer

Max-Sharpe / min-vol / target-vol under your stated caps, plus a drift-band rebalance check against the trigger you set

🛡️

Candidate Impact Preview

"Should I add this?" recomputes the book with the candidate and reports the delta — beta, volatility, 95% CVaR, correlation. A report, never a verdict

Search & Filter All 125+ Tools in the Interactive Directory →
Zero-Trust Architecture

Your Data. Your Machine. Period.

We built this for ourselves first. This is privacy-first, local-first AI investing research: every API key, every holding, and every memory fragment stays on your local disk. AI inference is performed via encrypted API calls to enterprise developer endpoints (AWS Bedrock, Vertex AI, Azure OpenAI and others), where your prompts are not used to train foundation models. Retention is a separate matter: most providers hold prompts briefly for abuse monitoring, and the window varies by provider — see the FAQ for the per-provider detail before you choose one, and verify it against your own provider's current terms, since these policies change.

Learn About Our Secure Credentials Keyring & Private API Support →
Quick Start

Deploy in Under 5 Minutes

Clone, install, and launch. The guided setup wizard handles the rest.

01

Clone & Install

One script installs all dependencies, creates your data directory, and migrates existing config.

git clone https://github.com/aymanyaq/cairniq.git && cd cairniq && ./install.sh
02

Configure Keys

The interactive Setup Wizard walks you through connecting your AI provider and brokerage accounts.

Settings → Commit Infrastructure Sync
03

Launch CairnIQ

Double-click the launcher or run the command. Your private intelligence console opens at localhost:8000.

./CairnIQ.command  or  CairnIQ.bat

Frequently Asked Questions

Here are a few quick answers to common questions about CairnIQ.

What LLM providers are supported?
Claude (Anthropic direct or via AWS Bedrock), OpenAI, Azure OpenAI / AI Foundry, Google Gemini via Vertex AI, and any OpenAI-compatible endpoint. The console uses a provider-agnostic abstraction layer, and the provider and model can be switched from Settings at runtime without restarting the server.
Is my financial data safe?
Your holdings, journal, chat history and memory graph live on your own disk in user_data/, and API keys go into your OS keychain — not into a cloud database. CairnIQ stores nothing on our servers, because there are no CairnIQ servers, and it ships no telemetry or analytics.

The honest caveat: answering a question about your portfolio means sending the relevant holdings to whichever LLM provider you configure. Those enterprise endpoints commit that your prompts are not used to train foundation models — that part is real, and it is the substantive difference from a consumer chatbot. But they are not zero-retention: most hold prompts for a limited abuse-monitoring window, from 30 up to 90 days depending on the vendor. See the per-provider breakdown, and confirm against your provider's current terms.
Read the Full FAQ & Technical Docs Page →

Latest Research & Insights

Articles on AI financial intelligence, portfolio analysis, and self-hosted engineering.

Read All Articles on Medium →

Take Cairn Intelligence

Your wealth intelligence. Your infrastructure. Your rules.

View on GitHub Request Support