Aewita vs. CoCounsel: what being a wrapper means
CoCounsel is a thin product over other people’s models and another company’s database. Aewita is one company, one model, one boundary. Here is why that difference matters.
A short history of CoCounsel
CoCounsel started as a product from Casetext, a legal research company known for its Parallel Search tool. In 2023, Thomson Reuters acquired Casetext for a reported $650 million and folded the team into the Westlaw business. Since then, CoCounsel has been repositioned as the AI layer on top of Westlaw: you buy Westlaw, you buy CoCounsel, and the product queries Westlaw’s case database to ground its answers.
The underlying model is not Thomson Reuters’. CoCounsel has publicly described itself as using frontier models from OpenAI and Anthropic. In practice, a CoCounsel query is routed through Thomson Reuters’ infrastructure and then out to a third-party LLM provider for inference.
The wrapper problem
Every AI product is built on top of something. The question is how many layers sit between the prompt and the answer, and who owns each layer.
A CoCounsel query crosses at least two organizational boundaries. The first is Thomson Reuters itself, which operates the product, the Westlaw retrieval layer, and the contractual relationship with the firm. The second is the LLM provider—OpenAI or Anthropic—which actually runs the reasoning. Both parties have enterprise agreements that restrict training and retention. Both are real companies with real security. But the prompt still travels across two separate trust domains.
Aewita is built differently. We host our own frontier reasoning model on infrastructure we control. There is no OpenAI call. There is no Anthropic call. The retrieval, the inference, and the citation verifier all run inside the Aewita boundary. For a matter governed by strict outside-counsel guidelines, that is the difference between one subprocessor listing and a chain of them.
Every organizational boundary a prompt crosses is a boundary your client’s guidelines will ask you about. Fewer boundaries, fewer questions.
Data provenance and what firms actually pay for
When a firm buys CoCounsel, the total cost is not one line item. It is three:
- A Westlaw subscription (large, per-attorney, annual).
- A CoCounsel subscription on top (separately priced).
- The implicit cost of routing privileged queries through OpenAI or Anthropic—not a dollar cost, but a governance cost.
The combined Westlaw-plus-CoCounsel line item is enterprise-priced. Thomson Reuters does not publish a seat price for CoCounsel alone because, practically, you can’t buy it alone. The product depends on a Westlaw subscription to retrieve authoritative case and statute text.
Aewita is a different transaction. One vendor. One subscription. One stack. $99 per month or $720 per year, no bundled dependencies. We built and license our own case and statute corpus: every U.S. case from 1665 to today, every federal statute, every state statute. No Westlaw license required.
Aewita’s alternative, in one diagram
If a CoCounsel query looks like: firm → Thomson Reuters → Westlaw retrieval → OpenAI/Anthropic inference → Thomson Reuters → firm, an Aewita query looks like: firm → Aewita retrieval → Aewita inference → Aewita citation verifier → firm.
One of those diagrams has five hops across two companies. The other has four hops inside one. We’re not claiming this makes Aewita magically better at every task. We’re claiming it makes privilege and subprocessor questions materially simpler.
Pricing reality
| Aewita | CoCounsel | |
|---|---|---|
| List price | $99/mo, $720/yr | Not published; enterprise |
| Required bundled product | None | Westlaw subscription |
| Inference provider | Self-hosted | OpenAI and Anthropic |
| Seat minimum | None | Yes |
| Trial | 14 days, full product | Sales-led |
| Annual commit | Optional | Typically required |
Coverage reality
On case law, both products have the depth. Westlaw’s corpus is famous, and it should be—it is one of the two canonical U.S. research databases. Aewita’s corpus covers every U.S. case from 1665 to today and every federal and state statute. A partner running a research query on either platform will find the controlling authority for a typical matter.
The difference is what the product does once it has the authority. CoCounsel is primarily a research and review tool. It answers questions about documents you upload, it summarizes depositions, and it retrieves Westlaw precedent. Aewita ships research plus drafting—full agreement, motion, and memo drafting tuned to 792 document types across 22 practice areas—plus playbooks, which are firm-specific drafting patterns the model is constrained to follow. Aewita also publishes an MCP endpoint at mcp.aewita.com, so your existing matter-management and DMS tools can connect to Aewita as a native data source.
CoCounsel tells you what the law says. Aewita tells you what the law says, drafts the document, and follows your firm’s house style while doing it.
Hallucination posture
Aewita publishes a measured hallucination rate: under 0.3% at a 95% confidence interval, across 800 structured queries. Every citation is checked by a verifier that resolves it against our own corpus before the answer is rendered. CoCounsel, like most legal AI products, speaks about hallucination control qualitatively rather than with a published confidence-interval number. Both products take hallucination seriously. Only one tells you the rate.
Who should pick which
If your firm has a deep existing Westlaw commitment, your workflows are built around Westlaw citations and KeyCite, and your partners want an AI layer that feels like a natural extension of a familiar research tool, CoCounsel will feel native. It is a reasonable product for firms already committed to the Thomson Reuters ecosystem.
If you want a single-vendor stack, a self-hosted model, transparent pricing, and a product that handles drafting and playbooks natively—without adding a second LLM subprocessor to your client disclosures—Aewita is the cleaner fit. See our full comparison for the side-by-side, and our security architecture for the exact boundary diagram.
One more honest note. CoCounsel has a bigger brand, a bigger sales team, and a parent company most GCs have heard of. That matters in procurement. What we offer in return is something procurement eventually wants too: fewer parties on the data-flow diagram, a number instead of an adjective on accuracy, and a price you can read on a webpage.
What changes when the model is yours
Owning the model is not just a security story. It changes how the product improves. When CoCounsel wants a new behavior—say, a better way to summarize expert-witness reports, or tighter handling of state-specific procedural rules—it has to coax that behavior out of a general-purpose model through prompting. The model itself is a moving target. OpenAI and Anthropic ship new versions on their own schedule, sometimes with quiet behavior changes. A prompt that worked last quarter can regress without warning.
Aewita can do the same prompt-engineering work, and we do. But because the model runs on our infrastructure, we can also do things a wrapper cannot: fine-tune on legal-specific corpora, freeze the model version for an entire firm so behavior doesn’t drift under their feet, and route the inference stack around new retrieval features without asking permission from a third party. That stability is why we are comfortable publishing a single hallucination number. A number that floats under you is not a number anyone should publish.
Rule 1.1 and technology competence
Comment 8 to ABA Model Rule 1.1 has required lawyers to keep up with the benefits and risks of relevant technology since 2012. For AI vendors, that means lawyers are now expected to ask specific questions: where does the inference happen, what are the retention terms, who is a subprocessor, and what is the measured error rate. A product that cannot answer those questions cleanly is a product that makes Rule 1.1 compliance harder, not easier.
CoCounsel can answer those questions, but the answers involve more parties. Aewita’s answers involve one. That simplicity is the entire point. The job of a legal AI vendor is not just to produce good answers. It is to produce answers the lawyer can defend in a deposition, in a bar complaint, or to a skeptical GC. We designed Aewita so that defense fits on one page.
If you are currently running CoCounsel and considering an alternative, the cleanest test is not a feature bake-off. Run the same three matters through both products for two weeks. Measure time-to-first-draft on your real agreements. Count the citations each product emits and spot-check a random 50 of them. Ask each vendor for their subprocessor list in writing. Then hand all of that to your information-security lead and let them make the call. We think Aewita wins that test on the merits. Start with the 14-day trial; no card required, no sales call, real product. The point of a short comparison is to make a long decision faster.
One vendor. One model. One stack.
14 days free. Real access to the real product.