Introduction
This guide outlines what you need to know to successfully implement Pramata's AI Agents. We will provide information on:
- The large language model (LLM) powering your GenAI Agents
- Deployment options available to you
- Estimating potential token processing costs based on use cases
Our goal is to equip you with the key information you require to effectively prepare for and derive maximum value from GenAI Agents.
Large Language Model: Anthropic Claude
Pramata's GenAI Agents are powered exclusively by Claude, specifically Claude Haiku 4.5 or later models. Claude has been chosen for its security, relevant knowledge, response quality and cost effectiveness.
Key Features of Claude:
-
Security:
- Ensures no data mixing across customers
- No persistence of customer data/training
-
Relevant Knowledge:
- Strong capabilities in legal domain and contract interpretation
- Extensive general business knowledge
-
Response Quality and Cost Effectiveness:
- Provides more consistent and easier to understand responses
- Offers reasonable token costs for high-quality outputs
- Delivers clear and actionable insights
Deployment Options
Pramata offers two deployment options for Claude:
| Pramata-Hosted Claude on Amazon | Customer-Hosted Claude on Amazon/Anthropic |
|
|
If you choose the Customer-Hosted option, you'll need to set up an Enterprise account with Anthropic or an Amazon Bedrock account. Here are links to help you get started:
- For Anthropic: Anthropic Claude API Setup Guide
- For Amazon Bedrock: Amazon Bedrock Claude Setup Guide
Our team is also available to assist you with the setup process and answer any questions you may have.
Estimating Potential Token Usage Costs
If you choose the Customer-Hosted LLM option, you'll need to consider token processing expenses. Exact figures will vary based on:
- The pricing model offered by Anthropic or Amazon Bedrock
- Specific GenAI Agents leveraged
- Volume of contract data processed.
Important Note: Token costs for LLMs are subject to change and have been trending downwards. The estimates provided here are based on current pricing, but actual costs in the coming months may be lower than these projections.
To determine appropriate budgets, use the following as a guide:
* Pricing as of November 2025
| Common AI Agents | Assumption (# of Pages / Records) | Tokens Used | Cost Per Interaction (Claude)* |
| Summarize Key Clauses | 21 | 15.2 | $0.02 |
| Analyze Drafts / Redlines | 8 | 5 | $0.01 |
| Account Review Memo | 30 | 54.8 | $0.07 |
| Reporting Analysis | 75 | 23.4 | $0.04 |
Estimated Monthly Projections (Claude Haiku 4.5):
| Claude Haiku 4.5 | # of Monthly Interactions | ||
| Low Usage | Medium Usage |
High Usage |
|
| Summarize Key Clauses | 250 | 500 | 1,000 |
| Analyze Drafts / Redlines | 100 | 250 | 500 |
| Account Review Memo | 50 | 100 | 250 |
| Reporting Analysis | 50 | 100 | 250 |
| Total Monthly Token Fees | $10 | $21 | $48 |
Conclusion
We understand that concepts around enterprise LLMs, token pricing, security and more can be disorienting at first. Our goal with this guide is to help you prepare for and extract value from Pramata’s GenAI Agents.
As you evaluate our GenAI Agents rollout, our team remains eager to collaborate however we can – whether assisting with LLM selection, estimating costs, or simply answering any other questions. We’re committed to ensuring you translate the “theoretical AI promise” into tangible business benefits.
If you choose the Pramata-hosted option, you can start benefiting from GenAI Agents without worrying about token costs or complex setups. If you prefer to manage your own Claude instance, we're here to help you navigate that process as well.
Please let us know how we can help accelerate your journey!
FAQs
Q: Does Pramata combine customer data with data from other clients?
A: Absolutely not. At Pramata, we ensure that your data remains exclusively yours. We do not mix or share your data with any other customer's data.
Q: Without using customer-specific data, how do Language Learning Models (LLMs) operate effectively?
A: LLMs, are equipped with extensive pre-existing legal knowledge. They don't require customer-specific data to function effectively.
Q: Does Pramata possess its proprietary LLM?
A: Pramata does not own a specific LLM. Our platform is architected to connect to industry-leading LLMs and at this time, we have selected Anthropic Claude as our preferred LLM due to it’s enterprise-grade security features, relevant knowledge, response quality, and cost effectiveness. Our focus and innovation lie in utilizing LLMs to tackle complex business challenges, providing substantial business value while maintaining the utmost security and privacy for your data. This approach includes three patents pending.
Q: If Pramata doesn't own the LLM, how can data security be guaranteed?
A: Pramata has opted to go with Claude, accessed through Amazon Bedrock. We selected this LLM because it benefits from AWS’ enterprise-level security infrastructure, including robust encryption, access controls, and compliance certifications. This integration ensures your data remains protected by industry-leading security measures. Customers also have the option of hosting Claude themselves and integrating their service URLs and passkeys to Claude inside our platform.
Q: How does Pramata optimize the costs associated with LLMs?
A: Pramata has developed innovative methods to extract maximum value from LLMs while minimizing token processing, making the costs of running an LLM service quite affordable. This cost efficiency is achieved without compromising the high level of business value provided.
Q: Do LLMs require high-quality data to function properly? How does Pramata address this need?
A: Indeed, the quality of input data is crucial for LLM effectiveness. Poor data quality increases the risk of errors and inaccuracies. Pramata addresses this by specializing in providing clean, non-redundant contract data, which is ideal for LLM processing and ensures more reliable outcomes.