Legal Infrastructure for AI

Your agent reads propositions,
not 30-page opinions

AgentLaw is legal research infrastructure where the primary consumer is an AI agent. Structured propositions with confidence scoring, authority hierarchy, and citation provenance — not document retrieval with an API wrapper.


Document retrieval is the wrong abstraction for agents

Existing platforms give you ranked document lists. Every agent must then read, parse, assess authority, and repeat — burning tokens and time on work that should be done once.

50K–300K
Tokens per research query
Document retrieval (Westlaw/Lexis)
vs.
< 500
Tokens per research query
Proposition lookup (AgentLaw)
Document-Oriented Proposition-Oriented
Unit of information A case or statute (document) A legal assertion with provenance
Search model Keywords → ranked document list Structured query → graph traversal
Authority ranking Human judgment at query time Machine-readable hierarchy in every response
Citation checking Visual signals (Shepard's / KeyCite) Citation network as a queryable graph
Currency "As of" dates on documents Temporal validity on every node + conflict detection
Jurisdiction Filters applied manually First-class jurisdiction hierarchy

Six layers of legal reasoning infrastructure

Each layer builds on the one below. Agents consume the layer they need — from raw propositions to full regulatory compliance graphs.

L1

Legal Knowledge Graph

Propositions as nodes. Typed edges (supports, contradicts, narrows, overrules). Authority hierarchy enforced at the data layer. Confidence scoring from four component signals.

L2

Reasoning Primitives

Authority lookup, statute search, graph traversal, conflict detection, and temporal queries — exposed as API calls, not left to the agent to figure out.

L3

Jurisdiction Intelligence

Hierarchy resolution, Golsen rule, preemption mapping, and regulatory-body jurisdiction. Binding vs. persuasive authority resolved before it reaches your agent.

L4

Temporal Legal State

Current-law resolution accounting for amendments and sunsets. Point-in-time queries. Change webhooks when the law evolves.

L5

Agent Workflow Primitives

Research contexts with cross-session continuity. Research chains for iterative refinement. Collaborative research so multiple agents avoid duplicating work.

L6

Compliance & Regulatory Graph

Obligation mapping, compliance posture scoring, and enforcement pattern analysis — structured for automated regulatory monitoring.


Authority hierarchy is enforced, not suggested

When authorities conflict, the higher source controls. This is programmatically enforced in every API response — your agent never needs to implement hierarchy logic.

1 Internal Revenue Code (IRC) — supreme
2 Treasury Regulations — controlled by Code
3 Tax Court & Federal Case Law
4 IRS Internal Revenue Manual — not law

Confidence Scoring

Every proposition carries a 0.0–1.0 confidence score computed from four signals: authority strength, recency, consistency, and novelty. Deterministic, auditable, re-scored when new cases arrive.

Conflict Detection

Propositions anchored to the same statute but at different authority levels are automatically flagged. Hierarchy resolution returns pre-grouped controlling, subordinate, and conflicting authorities.

Temporal Validity

Every proposition carries valid_from and valid_to dates. Superseded holdings don't silently pollute results — they're marked and traversable for historical research.


Built for agents, not attorneys

REST API with structured JSON responses. Every endpoint returns machine-readable authority metadata — no parsing required.

GET /v1/propositions/search?q=abuse+of+discretion
{ "query": "abuse of discretion", "results": [ { "id": "cdp-abuse-of-discretion-001", "proposition": "Abuse of discretion exists when a determination is arbitrary, capricious, or without sound basis in fact or law.", "authority_type": "holding", "confidence_tier": "controlling", "confidence_score": 0.915, "primary_citation": "Murphy v. Commissioner, 125 T.C. 301 (2005)", "statutory_anchors": ["IRC § 6330"], "edges": { "supports": 4, "narrows": 2 } } ] }

Proposition Search

Keyword and statute-based search returning structured propositions with confidence scores, not document snippets.

Graph Traversal

Follow typed edges between propositions: supports, contradicts, narrows, interprets, overrules. Build reasoning chains programmatically.

Authority Resolution

Given a statute section, get pre-resolved results grouped by authority level — controlling, subordinate, conflicting. No hierarchy logic needed.


Starting narrow, expanding methodically

Every proposition is domain-expert verified before entering the graph. We start where we have the deepest expertise and expand from there.

Now: CDP / Tax Controversy

248 verified propositions covering Collection Due Process hearings — deadlines, standards of review, collection alternatives, Tax Court procedure. Live at cdprights.com.

Next: Full Tax Controversy

Deficiency proceedings, innocent spouse, passport revocation, penalty abatement. Same architecture, expanding the graph.

Then: Tax Compliance & Beyond

Regulatory obligations, filing requirements, financial regulation. Each domain follows the same pattern: public data → knowledge graph → expert validation → API.


Per-query, not per-seat

Agents make thousands of queries. Per-query pricing aligns with how AI actually uses legal research.

Free

$0
Rate-limited
  • Proposition text previews
  • Keyword search
  • Topic browsing
  • Statute lookup

Enterprise

Custom
Dedicated infrastructure
  • Custom jurisdiction configurations
  • Research contexts with continuity
  • Change webhooks
  • Volume pricing

The Twilio of legal research, not the call center

AgentLaw is not a legal AI tool. It's the legal data infrastructure that legal AI tools query.

Company What they do How AgentLaw differs
Harvey AI assistant using legal databases Harvey is the agent; AgentLaw is the database
Casetext / CoCounsel AI on top of document stores Propositions, not documents
Westlaw API Document retrieval via API Different data model entirely
CourtListener Open legal documents Adds the structured knowledge layer

Connect your agent directly

AgentLaw exposes a remote MCP server. Connect any MCP-compatible agent to query the legal knowledge graph with 8 tools.

claude_desktop_config.json
{ "mcpServers": { "agentlaw": { "url": "https://agent-law.net/sse" } } }

8 Tools

search_propositions, get_proposition_details, get_by_statute, resolve_authority, graph traversal, topic search, review queue, list all.

SSE Transport

Standard MCP SSE protocol. Works with Claude Desktop, Claude Code, and any MCP-compatible client.


Ship legal AI that's actually fast

Stop burning tokens on document retrieval. Query structured legal knowledge directly.