CIO Roadmap · April 2026

Gáldr × Viska sjóðir

AI Intelligence Department — 6-Month Build & Operate Roadmap
Prepared for: Guðlaugur Steinarr Gíslason, CIO From: Gáldr ehf. Date: April 2026
01

What This Document Contains

Compiled research, testing results, and current development status from 9 formal investigations and 7 war room build sessions. This gives a concrete indicator of what we can and should build in the coming months.

Everything below is a proposal pending Gulli's approval. The next phase will consolidate these plans with the operational needs of Viska Crypto and Viska Macro as Gulli defines them.

What we need from Gulli:

Gulli's initial proposals (10 capabilities)

  1. AI-powered macro research pipeline (cross-asset, multi-source)
  2. Multi-agent investment analysis (autonomous surveillance agents)
  3. Real-time sentiment and narrative tracking (Grok + X integration)
  4. Automated investor communications and LP reporting
  5. Scenario modeling at scale (overnight deep analysis)
  6. AI-assisted due diligence on distribution partners
  7. AI-generated thought leadership content
  8. Open-source models for proprietary signal generation
  9. Meeting preparation and follow-up for fundraising
  10. Trading infrastructure with risk enforcement

These are the starting point. As the engagement progresses, goals adapt — new capabilities emerge, fund priorities shift, and we add deliverables at any phase boundary. The methodology section describes how we evaluate and prioritize new information.

02

Goals

Six capability milestones over six months. Each builds on the previous. Goals adapt to new capabilities and fund priorities as they emerge.

MonthGoalOutcome
1Macro Research IntelligenceFull corpus queryable (brokers + central banks + earnings). Daily brief. MCP access from any workspace.
2Multi-Agent Surveillance3 autonomous agents (crypto, macro sentiment, geopolitical risk) running 24/7. Alert pipeline.
3Trading War RoomPaper trading operational. 4 agents, 7 pipelines, risk framework enforcing, 7-page dashboard live.
4Investor CommunicationsLP letter pipeline, thought leadership (2–3/week), meeting intelligence (prep + follow-up).
5Scenario Modeling & SignalsOvernight scenario lab. Proprietary signal generator on Viska infrastructure.
6Distribution Intelligence & HandoffPartner surveillance (weekly scans). Full codebase handoff. Documentation. Training.
03

Methodology

Research → Evaluate → Decide → Build

Nothing is adopted without evaluation. Nothing is discarded without documented rationale.

Research conducted for this build: 9 formal investigations over 2 weeks, producing 374 lines of documented findings across 9 domains.

Research Inventory

#InvestigationDateKey FindingDecision
1n8n node capabilities (R2, SQLite, notifications, FinBERT)Apr 10S3 node supports R2 via custom endpoint. No built-in SQLite. ntfy for notifications.R2 adopted. ntfy as primary notification. FinBERT deferred.
2External API integration (Grok, Fed, ECB, Gemma)Apr 10Grok must use Responses API (Chat Completions deprecated). grok-4-1-fast is 10x cheaper. FOMC RSS confirmed.Responses API + fast model. CB Document Pipeline for Fed + ECB. Gemma 4 via Ollama overnight.
3MCP & CLI tool audit (11 services)Apr 124 new MCP servers: FRED (800k+ series), QuantOracle (63 tools), Ollama, SQLite. FRED + QuantOracle close quantitative macro layer.All 4 wired. This finding eliminated the case for TradingView MCP.
4TradingView MCP (2 projects)Apr 13Project 1: 5 kill factors (ToS, fragile CDP, 500MB, single-agent, irrelevant Pine). Project 2: backtesting useful, sentiment weak.Project 1 declined permanently. Project 2 selective for Phase 1D.
5MCP vs CLI architectureApr 7MCP costs 10–32x more tokens than CLI. Reliability: CLI 100% vs MCP 72% (25 runs).CLI-first strategy. MCP for specific guarantees only.
6Karpathy wiki patternApr 7Three-layer pattern (raw → wiki → schema). Index retrieval outperforms RAG at ~400K words.Adopted for knowledge base. No vector database overhead.
7NotebookLM pipelineMar 14Claude pipeline superior (99% vs 60% transcript success). NotebookLM audio is unique value.Not integrated except audio. Fabric prompt patterns noted.
8Security evaluationApr 12HIGH: @mastra-factory/config pulling from live branch tip. MEDIUM: 3 open vectors.Pinned to SHA c339bcc. 3 vectors tracked with enforcement actions.
9Mastra core upgradeApr 1218 releases analyzed (v1.10 → v1.24.1). 2 breaking changes, neither affects usage.Direct jump executed. Committed 148c93c.

Evaluation Criteria

CriterionThreshold
Terms of ServiceMust comply. Would we be comfortable if the vendor audited our usage?
Multi-agent compatibilityMust support simultaneous access. Single-seat tools rejected.
Failure modeMust fail loudly (error), not silently (wrong data). Silent failure = hard reject.
MaintenanceMust be actively maintained. Stale dependencies are liabilities.
Resource costMust fit within infrastructure budget without competing with core system.
ReplaceabilityMust be substitutable. Single points of failure are unacceptable.

Harness Engineering

Stanford and Tsinghua (March 2026): the same AI model delivers a 6x performance difference depending on orchestration. The model is a commodity. The orchestration layer is the asset.

  1. Subtract before you add. One company removed 80% of its tools and got better results.
  2. Build the layer that survives. Orchestration transfers across model generations. Investment compounds.
  3. Govern first, optimize second. Credential isolation, audit trails, scope boundaries, human approval gates — structural from day one.
04

Trading War Room

Four-layer system. No layer reaches into another's domain.

WAR ROOM 4 Claude Code agents (Operator, Analyst, Risk, Execution) AUTOMATION 7 n8n pipelines (data, signals, alerts, logging, reports) EXECUTION Alpaca Paper Trading API (long-only, human-approved) PRESENTATION React dashboard (7 pages) + FastAPI proxy + R2

Agents — Current Build

AgentRoleToolsScope
OperatorSession coordinator, final authorityAlpaca, FRED, OllamaSynthesizes all output. Approves/rejects orders. Writes thesis + session logs.
Technical AnalystChart analysis, trade ideasAlpaca, FRED, QuantOracle, SQLiteTradingView alerts, quant confirmation. Does NOT make trade decisions.
Risk ManagerPre-trade enforcementAlpaca, QuantOracle, SQLite8-item checklist. Soft veto — unsigned trades cannot proceed.
Execution AgentOrder validation, submissionAlpaca, SQLite4 hard gates: operator_approved, buy-only, market hours, size validated.

Sessions: Pre-market 08:30 ET, midday 12:30 ET, EOS 16:15 ET, plus alert-driven.

Order Flow

Analyst thesis
Risk qualification
Operator approval
Execution
Alpaca

Proposed Agent Team — Full Fund Operations

Every key function in a trading operation mapped to an AI agent role. Covers both Viska Crypto and Viska Macro. Gulli should review this roster and indicate which roles to prioritize, defer, or remove.

RoleFundFunctionStatusNotes / Questions for Gulli
Trading OperatorBothSession coordination, order approval, strategy thesisBuiltCurrently covers both funds. Split into fund-specific operators?
Technical AnalystBothChart analysis, pattern detection, trade ideasBuiltShared across funds or fund-specific views?
Risk ManagerBothPosition sizing, exposure limits, drawdown enforcementBuiltSame risk framework for both funds or separate parameters?
Execution AgentBothOrder assembly, validation, Alpaca submissionBuiltLong-only Phase 1. Crypto execution path differs from equities.
Macro ResearcherMacroDeep macro analysis — central banks, fiscal policy, ratesPlannedPrimary value for Viska Macro. Fed/ECB/BoJ pipeline ready.
Crypto AnalystCryptoOn-chain metrics, exchange flows, DeFi monitoringPlannedWhat on-chain data sources does the team currently use?
Sentiment AnalystBothNLP-based market sentiment from news + socialPlannedReplaces keyword counting. Grok for real-time X data.
Geopolitical ScannerBothRegulatory filings, sanctions, elections, policy shiftsPlannedWhich jurisdictions matter most for current holdings?
Quant StrategistBothBacktesting, signal optimization, factor analysisPlannedProprietary signal generation in Month 5.
Portfolio AnalystBothCross-fund exposure, correlation analysis, rebalancing signalsNewDoes the team currently track cross-fund correlation?
LP CommunicationsBothMonthly letters, performance attribution, quarterly reportsMonth 4What's the current LP reporting cadence and format?
Content ProducerBothThought leadership — newsletters, LinkedIn, SubstackMonth 4Continue Substack cadence? Target Nordic family offices?
Meeting IntelligenceBothPre-meeting prospect research, post-meeting follow-upMonth 4Family office, distribution partner, LP meetings
Partner IntelligenceBothDistribution partner surveillance — regulatory, AUM, personnelMonth 6Arena Wealth Management — other targets?
Knowledge SynthesiserBothWeekly reflection processing, skill library updatesPlannedNeeds 8 weeks of operational data first
Compliance MonitorBothRegulatory threshold monitoring, position limit checksNewSub-threshold AIFM limits — automated monitoring?

Live Data Integration — To Be Mapped

We need to connect the fund's live data streams to the intelligence platform. Pending input from Gulli.

Data StreamQuestionsImpact
Portfolio positionsWhere held? Format? API access?Real-time portfolio monitoring, risk enforcement
NAV / fund valuationFrequency? Source system?LP reporting, performance attribution
Trade execution recordsWhich brokers/exchanges? FIX protocol?Trade journal, P&L attribution
Crypto custodySelf-custody, exchange, institutional custodian?On-chain monitoring scope
Broker research accessDropbox (current). Other channels?Corpus expansion
LP registryWhere maintained? Format?CRM agent, meeting intelligence
Compliance dataAUM thresholds, investor qualification records?Compliance monitoring agent

Once mapped, these streams feed into the Viska Portal and the research pipeline — creating a unified intelligence layer across both funds.

Automation Pipelines

PipelineFunctionTriggerStatus
WatchdogPosition monitoring, L1/L2/L3 alertsEvery 5 min (08:00–20:00 ET)Live
Market DataAlpaca + FRED → JSON → R2ScheduledLive
Trade LogAgent trade files → SQLite + R2File triggerLive
Signal IntakeTradingView webhook → validate → enrich → SQLite → R2 → notifyWebhookLive
Weekly ReportQuantStats metrics, P&L attribution, narrativeFridaysLive
Grok NarrativeAI market narrative from macro + price dataScheduledBuilt
CB DocumentsFed FOMC + ECB rate processingScheduledBuilt

Risk Framework

RuleLimit
Max risk per trade1%
Max position size5%
Max sector exposure20%
Min risk/reward1.5:1
Max open positions8
Min cash reserve20%
Daily drawdown halt−5%
Peak-to-trough halt−10% (full review)
Consecutive losing days3 → mandatory reflection

Non-negotiable: Human approval on every order (schema-enforced). 90-day paper trading before live capital (hard gate). Phase 1: long-only, no options, no leverage, no crypto.

Viska Portal

A purpose-built web interface where all intelligence, trading data, and operational status is accessible to the entire team. Not a generic dashboard — every page is designed for the fund's workflows.

The portal already exists at boas.dev/viska/portal with live chat (Mímir), daily reports, and corpus intelligence. The trading war room adds a second portal surface at trading.boas.dev for market operations. Both move to Viska-owned domains at handoff.

Existing portal sections (live)

F1 Macro F2 TA F3 Corpus F4 News F5 Chat F6 Daily Reports F7 Custom Reports F8 Changelog

Trading portal pages (building)

PageData SourceContent
/overviewR2Portfolio value, P&L, thesis, regime assessment
/positionsFastAPI proxy (Alpaca live)Open positions, risk badges, drawdown gauge
/strategyR2Current thesis, watchlist status
/analyticsR2 + FastAPIEquity curve, Sharpe, win rate, P&L attribution
/logR2Complete trade journal with decision chains
/alertsR2Alert timeline, L1/L2/L3 hierarchy
/marketTradingView widgetsLive embeddable charts for 7 instruments
/roadmapR2Development progress — shipped, in progress, coming next

The portal expands as capabilities come online. Months 4–6 add sections for LP communications, content pipeline status, scenario analysis results, and partner intelligence feeds.

Slack Integration & Scheduled Reports

The intelligence layer is accessible from two surfaces: the portal (visual, team-wide) and Slack (conversational, immediate).

CapabilitySurfaceTrigger
Mímir research queriesSlackOn-demand — any team member asks a question
Daily intelligence briefSlack + PortalScheduled — delivered every morning
Weekly performance reportSlack + PortalScheduled — every Friday, QuantStats + P&L + narrative
L2/L3 alertsSlack (ntfy)Event-driven — position monitoring triggers
TradingView signal intakeSlack (ntfy)Event-driven — webhook fires on alert conditions
Grok narrative briefingSlack + PortalScheduled — AI-generated market narrative
Report Studio outputSlackOn-demand — /report generates formatted deliverable

Reports run on n8n schedules. Intelligence arrives where the team already works.

05

Surveillance Agents

AgentDomainOutputCadence
CryptoOn-chain metrics, exchange flows, funding ratesAnomaly reports, whale alertsContinuous
Macro SentimentFX, rates, commodities, equitiesConviction scoring, divergence flagsDaily + continuous
Geopolitical RiskRegulatory, sanctions, electionsRisk event alerts, jurisdiction monitoringContinuous
06

Intelligence & Communications

AgentDomainOutputMonth
MímirBroker research corpusSourced answers, cross-broker synthesisLive
Daily BriefOvernight synthesisThemed morning report, high-impact flagsLive
Content EngineThought leadership2–3 pieces/week (IS + EN), newsletters, LinkedIn4
CRM AgentLP relationshipsMeeting prep, follow-up drafts, relationship queries4
Scenario LabOvernight deep analysisScenario reports, probability, portfolio impact5
Signal GeneratorProprietary signalsBacktestable signals from Viska framework5
Partner IntelligenceDistribution surveillanceWeekly: regulatory, AUM, personnel, news6
07

Tool Stack

Adopted

ToolTypeCapabilitySelection Rationale
Alpaca MCPAgent MCP70+ tools: account, orders, positions, market dataOfficial API, paper + live, no ToS concerns
FRED MCPAgent MCP800,000+ macro data seriesOfficial Fed data, maintained, discovered in MCP audit
QuantOracleAgent MCP63 deterministic tools: Sharpe, Sortino, GreeksProduction-grade, 1000 free/day, discovered in audit
Ollama MCPAgent MCPLocal inference (Gemma 4)Own hardware, no API costs, synthesis tasks
SQLite MCPAgent MCPDirect SQL to trading.dbZero-latency session access to logs/signals
TradingView WebhooksSignal deliveryAlert-driven signal intakeOfficial, supported, no scraping
Cloudflare R2StorageStatic JSON for dashboardZero-cost reads, n8n writes via S3 node
n8nAutomation7 pipelinesSelf-hosted, auditable, no vendor lock-in
alpaca-py v0.43.2PythonAlpaca SDKOfficial (replaces deprecated alpaca-trade-api)
pandas-taPythonTechnical indicatorsConfirmation layer for analyst
PyPortfolioOptPythonPosition sizing (mean-variance, HRP)Risk manager tool
Riskfolio-LibPythonTail-risk (CVaR)Risk manager tool
QuantStatsPythonPerformance tearsheetsWeekly report generation

Declined

ToolDecisionKill Factors
TradingView Chart MCPDeclined permanently(1) ToS violation — headless browser scraping. (2) Fragile — CDP 889+ issues. (3) 500MB+ RAM. (4) Single-agent lock. (5) Pine Script irrelevant.
Browser-automation integrationsPolicy: never adoptAny tool automating a browser against a service's web UI creates fragile, ToS-violating dependencies.
Autonomous order executionDesign principleHuman approval is permanent. Schema-enforced. Not a Phase 1 limitation.
NotebookLM integrationNot integratedClaude pipeline superior: 99% vs 60% transcript success. Audio noted for future.

Selective / Deferred

ToolDecisionRationale
TradingView MCP (Python)Selective — Phase 1DPure Python, no browser automation. Backtesting + screening + multi-timeframe. 8 of 30 tools installed.
FinBERTDeferred — Phase 1DRequires FastAPI sidecar. Manual news monitoring until deployed.
Lean CLI (QuantConnect)Evaluate — Phase 1DBacktesting alternative. Requires Docker.

Architecture-Shaping Research

FindingImpact
FRED MCP + QuantOracle close the quantitative macro layerEliminated strongest argument for TradingView MCP adoption
MCP costs 10–32x more tokens than CLI (25 runs)CLI-first strategy. Selective MCP tool exposure.
GitHub MCP burns ~55K tokens for 43 toolsNever install 30+ tools unfiltered.
Grok must use Responses APIWorkflow uses correct API path from day one
Index retrieval outperforms RAG at ~400K wordsKnowledge base uses index retrieval, no vector DB
n8n S3 node supports R2 via custom endpointR2 as zero-cost static data store
@mastra-factory/config pulling from live branch tipPinned to SHA. Supply chain risk eliminated.
08

How We Handle Scope Changes

New information arrives constantly — new models ship, fund priorities shift, market conditions change, the team discovers capabilities they didn't know they needed.

At every phase boundary (monthly), we:

  1. Assess — what changed since the last plan? New tools? New fund priorities? Capabilities that exceeded or fell short?
  2. Evaluate — apply the same six-criteria framework to any new proposal.
  3. Reprioritize — adjust next month's goals to reflect actual needs, not the original plan.
  4. Document — every change enters the research-to-decision pipeline. The rationale exists for future reference.

New deliverables can be added at any phase boundary. The CIO's original 10 plans are the starting point, not the ceiling.

09

Security

PrincipleImplementation
Credential isolationKeys never enter AI reasoning layer. Runtime injection via sandbox. AI uses credentials, cannot see them.
Scope boundariesEach agent has defined domain, enforced mechanically. Research ≠ financial data. Communications ≠ trades.
Audit trailsEvery decision, action, data access logged and traceable.
Human authorityNo consequential action without approval. Permanent.
Browser securityCredentials never reach dashboard. FastAPI proxy handles auth server-side. Read-only frontend.
10

Ownership

All code on Viska GitHub. Knowledge base on Viska Supabase. Claude subscription on mimir@viskasjodir.is. Dashboard on Viska domain. Nothing locked in.

11

Capability Matrix

CapabilityM1M2M3M4M5M6
Research corpus query
Daily intelligence brief
Central bank sources
MCP workspace access
24/7 surveillance agents·
Sentiment tracking·
Paper trading + risk enforcement··
Live dashboard (7 pages)··
TradingView signal intake··
63 quant analysis tools··
LP letter automation···
Thought leadership (2–3/wk)···
Meeting intelligence···
Overnight scenario analysis····
Proprietary signal generation····
Partner surveillance·····
Full ownership + handoff·····