Ertas + OpenClaw for Law Firms

The Challenge

OpenClaw's ability to triage emails, review documents, extract contract terms, and draft correspondence makes it a natural fit for legal workflows. But law firms operate under constraints that make the default cloud API architecture untenable.

Attorney-client privilege protects communications between lawyers and their clients from disclosure. When a lawyer uses OpenClaw with a cloud API to review a privileged document, that document's contents are transmitted to OpenAI or Anthropic's servers as prompt input. This constitutes disclosure to a third party — and courts have increasingly scrutinised whether such disclosures waive privilege. A single API call containing privileged material could, in theory, compromise privilege for an entire matter.

Beyond privilege, law firms face regulatory requirements around data handling. The Australian Privacy Act, GDPR (for firms with EU clients), and various bar association rules impose specific obligations on how client data is stored, processed, and transmitted. Routing client data through commercial AI APIs creates compliance complexity that most firms cannot justify — especially when the opposing counsel could argue that the firm's AI workflow constituted an unauthorised disclosure.

The irony is that the tasks OpenClaw excels at — document review, email classification, clause extraction, memo drafting — are exactly the tasks that consume the most paralegal and junior associate time. The productivity gain is real. The compliance risk just needs to be eliminated.

The Solution

Ertas enables law firms to deploy OpenClaw with fine-tuned local models that keep all data on the firm's infrastructure. The architecture is straightforward: fine-tune a model on the firm's legal domain data in Ertas Studio, export as GGUF, deploy via Ollama on a firm-controlled server, and configure OpenClaw to use the local endpoint exclusively.

The fine-tuned model delivers better results than a generic cloud API on legal-specific tasks because it has been trained on the firm's actual work product. A model fine-tuned on 6 months of contract review annotations learns the firm's specific risk criteria — what constitutes an unfavourable indemnification clause, what liability caps are acceptable for different deal sizes, which jurisdiction-specific terms require flagging. A generic GPT-4o applies general legal knowledge; a fine-tuned model applies the firm's specific standards.

For multi-practice firms, Ertas's LoRA adapter system allows practice-group-specific customisation on a shared base model. The corporate team's adapter is trained on M&A documents and due diligence checklists. The litigation team's adapter is trained on discovery responses and motion drafts. The IP team's adapter is trained on patent specifications and prosecution history. Each practice group gets a specialised AI agent without deploying separate models.

Key Features

Example Workflow

A mid-size commercial law firm in Sydney deploys OpenClaw to accelerate contract review for its corporate advisory practice. The firm's paralegals currently spend 8-12 hours per week reviewing vendor agreements, flagging unfavourable terms, and extracting key dates and obligations into a matter management system. The firm exports 400 annotated contracts from their document management system — each with paralegal-marked clauses (flagged/acceptable) and extracted data fields. This dataset is uploaded to Ertas Studio, where a Llama 3.3 8B base model is fine-tuned using LoRA (rank 16, 3 epochs). The resulting model achieves 90% accuracy on clause flagging against a held-out test set — compared to 72% from a prompt-engineered GPT-4o using the same criteria. The model is exported as GGUF (Q5_K_M quantisation) and deployed on a Mac Studio M2 Ultra in the firm's server room. OpenClaw is configured to use the local model exclusively — no cloud API fallback. The firm's contract review OpenClaw agent now reads incoming contracts via email, extracts key terms into structured JSON, flags unfavourable clauses with explanations referencing the firm's criteria, and drafts a review memo for the supervising partner. Paralegal review time drops from 90 minutes to 15 minutes per contract, with the paralegal focused on verifying the AI's flagged items rather than reading every clause from scratch. Total deployment cost: the Mac Studio hardware (AU$5,500 one-time) plus Ertas subscription. No ongoing API costs. No privileged data leaving the building.