Competitive analysis
Kervalt vs ChatGPT & Claude
In every legal AI sales call, the real incumbent is the $20-a-month DIY default. ChatGPT and Claude are familiar, zero-procurement, and genuinely good at drafting. But for legal work they share the same structural holes: no legal databases, no citation verification, no matter organization, prompts-not-workflows, client data leaving to the vendor's cloud, and zero firm-specific governance. This comparison works differently. Because the strategy does.
Investigation date: October 22, 2026 · Sources: openai.com and anthropic.com pricing pages
The strategy
Don't compete with the model. Absorb it
Kervalt's Rust server can keep OpenAI, Anthropic, or a local model behind the same gates. The model is a replaceable engine; the capability map, Source, and SOP are the product. Claude's strengths even map to ours: its long-context document analysis is our multi-doc run, its care and restraint is our verifier role.
- ChatGPT is a brilliant lawyer with no memory of your firm, no sources it trusts, and no supervisor. Kervalt gives it all three.
- Claude is our favorite engine too. It's inside. The difference is what surrounds it: sources it must cite, gates it must stop at, and a map of your firm it must work through.
- Every ChatGPT seat in a firm doing client work is a compliance incident waiting for our audit-log pitch.
Gap backlog
What the DIY default forces us to nail
P1: this quarter
Model flexibility
-
Model-provider abstraction
OpenAI / Anthropic / local (Ollama) behind one interface, with per-capability model pinning.
P2: amplifiers
Migration paths
-
Import your ChatGPT prompts as Skills
A migration path that turns a firm's prompt collection into governed Kervalt skills.
The field guide
Who you actually fight per deal
| When the buyer says… | You're up against | Your winning line |
|---|---|---|
| “We use Libra or Alice already” | Legal workspace | They answer. Kervalt operates: gates, scoped agents, your map. |
| “Harvey quoted us” | Enterprise AI | Per workspace, your server, no seat minimums, and our AI stops where you say stop. |
| “We just use ChatGPT / Claude” | The DIY default | Same engine, plus sources, gates, and an audit trail your insurer will love. |
Moats
What none of them can copy without rebuilding
- The capability map as runtime: your firm, executable
- Policy-fired gates with human tokens
- Capability-scoped, versioned Source
- A self-hosted single binary
Merged backlog
Every gap, one priority list
- P0
- Review/diff mode · visible citations · legal connectors (demo) · DOCX export · case-analysis graph
- P1
- SOC 2 story · model-provider abstraction · Discovery surface · compliance pack
- P2
- Per-case threads · Word add-in · prompt import
Evaluating Kervalt against something else? We'll investigate it and publish the comparison.
NEXT STEP
Ready to see the full picture?
See how your team finds hidden risks and decides with confidence, in minutes.