Legal workspace benchmark: Relex vs Google Workspace and other workspaces
Google Workspace keeps your documents and email. Relex keeps the legal matter ready for your next authorized agent. Choose the layer you need—not another model to commit to.
Named alternatives: what each layer does
| Workspace | Documented role | What to evaluate for legal work |
|---|---|---|
| Google Workspace | Email, files and collaborative documents, with AI assistance [1] | A reusable legal matter record alongside your existing Docs and Drive |
| Microsoft 365 | Office collaboration; agents can use SharePoint and connected knowledge [2] | Matter structure and handover across your chosen assistants |
| Box | Governed content; external AI access through Box MCP [3] | Legal workflow and saved decisions beyond document retrieval |
| Notion | Pages and databases; external tools can read and write through Notion MCP [4] | Legal entities, protected identities and professional handover |
| Relex | Intake, knowledge, research, matters, agreements, signatures and billing [5] | Reuse saved facts, sources and progress through supported connectors |
Product-fit comparison, not a performance ranking. Plans, integrations and setup matter. See the comparison for sources and a repeatable evaluation protocol.
Benchmark protocol: prove the handover, not the slogan
Use one synthetic matter, two authorized assistants and a separate unauthorized account. Record product/plan, configuration, model versions, date and every manual step. Run the same tasks three times per setup. Do not use client data.
- Context continuity
- Agent A saves three facts, two source links and a conclusion. Agent B retrieves all six without re-uploading or retyping. Record omissions and manual steps.
- Access boundary
- A second tenant cannot read the matter. Revoke access and retry. Record denial, token expiry and revocation behavior—not just the UI disconnect.
- Identity exposure
- Plant a synthetic name and email. Inspect outbound tool payloads and logs for raw values. Record every permitted disclosure and residual identifying facts.
- Token cost
- Compare full-context and targeted retrieval with the same questions and required answers. Count setup, input/output tokens, tool calls, cache, retries and service charges. Report quality alongside cost.
- Custom platform
- Build intake → matter → saved result → second-agent read. Record integration time, custom code, access controls and which steps used mocks.
- Professional handover
- Find a matching published profile, prepare an invitation, then require user approval. Record what was shared. No eligible profile is a result—not a successful referral.
Status: documentation comparison reviewed 25 September 2026. Cross-vendor execution results are not yet published; no winner, savings percentage or legal-quality score is claimed. Published by Relex, not an independent testing lab.
Local encryption check · 5 runs
The compiled browser crypto recovered the same synthetic identity after serializing encrypted cloud records; wrong passwords, wrong keys and modified ciphertext were rejected in all five runs. Cloud storage and a second device were simulated. This does not measure host connections, document detection, token savings or competitors.
Download results and limitations JSON · v1