AIP (AI Platform) — Complete Overview & Architecture

What is AIP?

Palantir AIP (AI Platform) is the AI layer built on top of Foundry and the Ontology. It enables organizations to deploy production-grade AI agents, copilots, and AI-powered workflows that act on real enterprise data — all with built-in governance, auditability, and security.

AIP was announced at Palantir Forward 2023 and has become Palantir's fastest-growing product, driving significant commercial revenue.

AIP Architecture Components

1. AIP Logic

AIP Logic is a serverless function environment where TypeScript functions call LLMs and interact with the Ontology. Functions can:

• Retrieve Ontology objects as context for LLM prompts
• Apply Ontology Actions to update business data
• Chain multiple LLM calls in a pipeline
• Return structured outputs consumed by Workshop, Copilot, or external APIs

// AIP Logic Function Example
import { Function, Context } from '@osdk/aip';
import { MissionPlan, ThreatAssessment } from './ontology-objects';

export const analyzeThreatAndSuggestAction = Function({
  name: 'analyze-threat',
  description: 'Analyzes a threat assessment and suggests a mission response',
  parameters: {
    threatId: { type: 'string', description: 'Threat assessment object ID' },
  },
  
  async execute({ threatId }, context: Context) {
    // 1. Fetch Ontology object — real, live data
    const threat = await context.ontology.ThreatAssessment.get(threatId);
    
    // 2. Call LLM with grounded context
    const analysis = await context.llm.complete({
      model: 'gpt-4o',
      system: 'You are a military planning assistant. Analyze the threat and recommend action.',
      user: `Threat Data: ${JSON.stringify(threat)}

Provide: severity rating, recommended response, estimated timeline.`,
      outputFormat: { severity: 'string', response: 'string', timeline: 'string' },
    });
    
    // 3. Apply an Action to record the analysis
    await context.ontology.actions.recordThreatAnalysis({
      threat,
      aiAnalysis: analysis.response,
      severity: analysis.severity,
    });
    
    return analysis;
  },
});

2. AIP Copilot

Copilots are AI assistants embedded into Workshop applications. They use AIP Logic functions as tools, allowing users to ask natural-language questions and trigger business operations.

Copilot Anatomy:

• Trigger: User types a message in a Workshop Copilot widget
• Intent Detection: LLM identifies which AIP Logic function to invoke
• Function Call: AIP executes the function with extracted parameters
• Response: Structured output formatted as a natural-language response

3. AIP Studio

AIP Studio is a visual agent builder — drag and drop LLM nodes, data retrieval steps, actions, and conditional branches to create multi-step AI workflows without code.

Studio Node Types:

• LLM Node: Call any configured LLM model
• Ontology Query Node: Retrieve objects by filter
• Action Node: Apply an Ontology action
• Branch Node: Conditional logic based on LLM output
• Loop Node: Iterate over a list of objects

4. Function Repository

TypeScript functions that extend the Ontology — callable from AIP Logic, Workshop widgets, Copilot, and external APIs via the OSDK.

Supported LLMs in AIP

AIP Security & Governance

Critical for enterprise adoption:

• All LLM inputs/outputs are logged and auditable
• Markings are enforced — LLMs only see data the user can access
• No training on customer data (explicit Palantir policy)
• RBAC controls which users can access which AIP functions
• PII/PHI masking can be applied before LLM calls

Real-World AIP Use Cases

1. Defense: Autonomous mission planning with threat assessment data from the Ontology
2. Healthcare: AI-assisted clinical decision support with live patient data
3. Finance: Automated regulatory compliance review of contracts and filings
4. Supply Chain: Predictive disruption alerts with AI-generated mitigation plans
5. Government: Benefits eligibility determination with auditability for appeals

Getting Started with AIP Logic

# Install Palantir CLI
npm install -g @osdk/cli

# Authenticate to your Foundry stack
osdk auth login --foundry-url https://your-stack.palantirfoundry.com

# Create a new Function Repository
osdk function create my-aip-functions

# Develop locally with live Ontology access
osdk function dev

AIP vs. Direct LLM API Calls