# AgentLaw — Legal Research Infrastructure for AI Agents

> Structured legal propositions, not documents. Machine-readable legal knowledge with authority hierarchy, confidence scoring, and citation provenance. Built for agent consumption.

AgentLaw is legal research infrastructure where the primary consumer is an AI agent. The unit of information is a structured legal proposition — not a document. A proposition lookup returns the relevant law with citations, authority ranking, and graph context for a fraction of the token cost of document retrieval.

## Why AgentLaw Exists

Existing legal research platforms (Westlaw, Lexis) give document retrieval with an API. Every agent must then:
1. Retrieve full case documents (5,000-20,000 tokens each)
2. Parse text to extract relevant holdings
3. Assess authority weight, jurisdiction scope, and temporal validity
4. Repeat for every query

AgentLaw pre-structures the knowledge. A proposition lookup costs hundreds of tokens, not hundreds of thousands.

## What a Proposition Contains

Each proposition node carries:
- Unique ID (e.g., `cdp-abuse-of-discretion-001`)
- Proposition text (structured legal assertion)
- Authority type: holding, dicta, statutory_text, reg_interpretation, procedural_rule, agency_guidance
- Confidence tier: controlling, persuasive, unsettled, dicta_only
- Confidence score: 0.0-1.0 (computed from authority_strength, recency, consistency, novelty)
- Uncertainty type: settled, unsettled (circuit split), undeveloped (no authority)
- Statutory/regulatory anchors (IRC sections, Treasury Regulations, Tax Court Rules)
- Source citations with pinpoints
- Jurisdiction scope
- Temporal validity (valid_from, valid_to)
- Typed graph edges to related propositions (supports, contradicts, narrows, interprets, overrules, etc.)

## Authority Hierarchy (Enforced Programmatically)

When authorities conflict, the higher source controls — always:
1. Internal Revenue Code (IRC) — supreme
2. Treasury Regulations — controlled by Code
3. Tax Court and Federal Case Law — controlled by Code and Regs
4. IRS Internal Revenue Manual (IRM) — NOT law; useful for procedure only

This hierarchy is enforced in every API response. Agents never need to implement hierarchy logic.

## Architecture Layers

- L1: Legal Knowledge Graph — propositions as nodes, typed edges, authority hierarchy, confidence scoring
- L2: Reasoning Primitives — authority lookup, statute search, graph traversal, conflict detection, temporal queries
- L3: Jurisdiction Intelligence — hierarchy resolution, Golsen rule, preemption mapping
- L4: Temporal Legal State — current-law resolution, point-in-time queries, change webhooks
- L5: Agent Workflow Primitives — research contexts, research chains, collaborative research
- L6: Compliance & Regulatory Graph — obligation mapping, posture scoring, enforcement patterns

## MCP Server (Remote)

Connect any MCP-compatible agent to the AgentLaw knowledge graph:

- **SSE Endpoint**: `https://agent-law.net/sse`
- **8 tools**: search_propositions, get_proposition_details, get_propositions_by_statute, get_related_propositions, list_all_propositions, get_propositions_by_topic, resolve_authority, get_review_queue

Claude Desktop config:
```json
{
  "mcpServers": {
    "agentlaw": {
      "url": "https://agent-law.net/sse"
    }
  }
}
```

## API Access

REST API with two access tiers:

**Free tier** — text previews, no confidence scores:
- `GET /v1/propositions` — list all (previews)
- `GET /v1/propositions/search?q=...` — keyword search
- `GET /v1/topics/{topic}` — topic browse
- `GET /v1/statutes/{anchor_type}/{section}` — statute lookup

**Paid tier** — full details, scores, graph, hierarchy:
- `GET /v1/propositions/{id}` — full proposition with citations, anchors, edges
- `GET /v1/propositions/{id}/related` — graph traversal
- `GET /v1/statutes/{anchor_type}/{section}/resolve` — hierarchy-resolved lookup
- `GET /v1/review-queue` — low-confidence propositions
- `POST /v1/candidates` — submit new case/revision to candidate queue
- `POST /v1/candidates/{id}/process` — process through re-scoring pipeline

## Current Coverage

- **Live now:** 248 verified propositions on Collection Due Process (CDP) hearings — deadlines, standards of review, collection alternatives, Tax Court procedure
- **Demo:** [cdprights.com](https://cdprights.com) — browse all propositions with confidence scores
- **Expansion:** Tax controversy (deficiency, innocent spouse, penalties) → tax compliance → financial regulation → multi-domain

## Machine-Readable Resources

- [Full Platform Description](https://agent-law.net/llms-full.txt): Complete architecture, API details, data model, and competitive positioning
- [CDP Knowledge Base](https://cdprights.com/llms-full.txt): All 248 CDP propositions with citations, anchors, and relationships
- [CDP Propositions JSON](https://cdprights.com/data/propositions.json): Structured proposition data
- [CDP Graph Edges JSON](https://cdprights.com/data/edges.json): Typed relationships between propositions

## Pricing

- Per-query pricing (not per-seat) — aligns with agent usage patterns
- Free tier: text previews, keyword search, topic browse, statute lookup
- Pro tier: full propositions, confidence scores, graph traversal, authority resolution
- Enterprise: custom jurisdiction configs, research contexts, change webhooks

## Key Differentiator

AgentLaw is not a legal AI tool. It's the legal data infrastructure that legal AI tools query. The Twilio of legal research, not the call center.

## Contact

- Website: [agent-law.net](https://agent-law.net)
- CDP Demo: [cdprights.com](https://cdprights.com)