Intelligence, ready to integrate

Legal intelligence.
Built into your next idea.

The reasoning, research, and tools behind AnrakLegal. Ready for your applications, agents, and workflows.

Request a demo Start building

One platform. Three ways to build.

AnrakLegal Research / 2026

Legal reasoning, measured.

92.44%

LegalBench macro score · Nine-task sample

126 / 126

Items with complete receipts

1,948

Recorded upstream calls

Explore the evaluation

Choose your starting point

What are you building?

Start with the capability you need.
Go deeper when you're ready.

/00

Jolly ANRAK

anraklegal/jolly

A multimodal system made for everything.

Manual API keys · Automatic app access

Get Jolly access
/01

Model API

Build with reasoning

Give your application a legal reasoning engine. Use Conductor for legal research or Ensemble for general reasoning.

OpenAI-compatible · Streaming · Durable tasks

Explore the models
/02

Data API

Build with legal data

Bring judgments, statutes, citation treatment, and global legal sources into your own research pipeline.

REST endpoints · Structured results · Usage metering

Explore the Data API
/03

MCP & integrations

Connect your assistant

Bring legal research to your existing AI workspace. Connect through MCP or configure Custom GPT Actions.

MCP · OAuth · Agent tool gateway

Connect with MCP

Research spotlight · August 2026

Stronger reasoning.
Evidence behind it.

Our receipt-verified LegalBench evaluation scored ahead of the standalone systems in the report's comparison. Every reported AnrakLegal item carried a complete execution record.

Read the research report (PDF, opens in a new tab)

AnrakLegal Research · 14 pages · PDF

LegalBench · Macro score across nine tasks
AnrakLegal92.44%
DeepSeek V4 Pro90.45%
GPT-5.589.63%

Source: report, Tables 1 & 3. AnrakLegal: 126 items (14 per task); standalone systems: 50 per task. Same harness and task set, different sample sizes. Retrieval disabled.

How to read these results

This is AnrakLegal Research's August 2026 evaluation of a seeded, stratified US-law sample. The 92.44% figure is an unweighted macro average across nine tasks, not a full-suite LegalBench score. No confidence intervals are reported for this sample.

The report also covers CUAD, ContractNLI, MAUD, and IL-TUR CJPE. MAUD scored 47.10% against a 47.80% baseline; ContractNLI has no claimed comparison baseline. Execution receipts document process completion, not guaranteed answer correctness. Mean LegalBench item latency was 86.73 seconds.

From idea to first request

A small start. A lot of possibility.

Your first legal answer.

Send a question. Receive a researched answer through a familiar chat completions endpoint.

  1. 1

    Create a scoped key

    Select Model API and Indian case law permissions in the console.

  2. 2

    Prepare your account

    Enable AnrakConductor in Labs and add Model API credits.

  3. 3

    Make your first request

    Set ANRAK_API_KEY in your environment, then run the example.

Full Model API guide
terminal
curl https://anrak.legal/api/v1/chat/completions \
  -H "Authorization: Bearer $ANRAK_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "anrak-conductor",
    "messages": [{
      "role": "user",
      "content": "Explain arbitration in India."
    }]
  }'

cURL

JSON response · Read the answer at choices[0].message.content

Keep building

Everything else, in its place.