Key facts
| Access control | Scoped API keys with rotation; enterprise SSO and RBAC options |
| Matter separation | Separate keys and retrieval indexes per matter or client |
| Deployment | Cloud, VPC, on-prem or air-gapped for confidential client work |
| Auditability | Request, model and response logging for matter review trails |
| Data grounding | Embeddings and RAG are live for precedents and firm know-how |
| Models | 30+ models behind one OpenAI-compatible API |
| Pricing model | Flat monthly self-serve plans; no per-token billing on self-serve |
| Endpoint roadmap | Files, batch and fine-tuning are coming soon |
TL;DR
- Treat privilege as the top classification axis, not an afterthought.
- Enforce matter walls in keys and retrieval indexes, not just policy text.
- Keep firm know-how and client files in separate indexes.
- Log requests, sources and reviewing lawyers for every client-facing output.
- Verify citations; models can produce plausible but wrong authority.
How it works, step by step
- Inventory AI use cases across research, drafting, review and business development.
- Classify content by privilege, matter and client confidentiality level.
- Choose deployment per class: cloud, VPC, on-prem or air-gapped.
- Issue per-matter or per-client keys and separate retrieval indexes.
- Define log fields and retention: request ID, model, sources, output, reviewing lawyer.
- Approve a model allow-list and require citation verification before filing.
- Apply legal hold rules to prompts, outputs, logs and retrieval indexes.
Try it yourself
Privilege, matters and data classification
Law firm content divides into public law and commentary, firm know-how, matter files, and privileged communications. Each class needs its own handling rule, and matter files need client-level separation because privilege follows the client relationship, not the practice group.
Write the classification down before any pilot. A research assistant grounded in public materials is low risk; a workflow that reads matter files is not, and the two should never share a retrieval index.
Keys, matter walls and least privilege
Issue a distinct API key per application and, where feasible, per matter or client, then keep indexes separated so a retrieval call cannot cross a matter wall. Store keys in a secrets manager, rotate on a schedule, and revoke them when a matter closes or a lawyer leaves. Enterprise SSO and RBAC options keep console access aligned with the firm's conflicts and personnel processes.
Never place client names or privileged excerpts in prompts when retrieval can supply only the passage a task needs.
Residency, retention and legal hold
Decide where processing happens and how long records live. Region selection covers many residency needs; VPC, on-prem and air-gapped deployment keeps confidential material inside the firm or client environment. Retention must respect engagement terms and legal hold, which means deletion of prompts, outputs, logs and indexes must be technically enforceable, not aspirational.
Log enough to reconstruct an output: request ID, model and version, retrieved source identifiers, output and the reviewing lawyer. See AI audit logs for a schema.
Model governance and professional accountability
Keep an approved model list with evaluation evidence and re-test when versions change. Require citations from approved sources and independent verification, because a confident answer is not the same as correct authority. A qualified lawyer remains accountable for advice and filings, so the checklist should name the reviewer and state that AI output is a draft until verified.
Honest comparison
| Control area | Plugsky capability | Common gap | Owner |
|---|---|---|---|
| Matter separation | Separate keys and indexes per matter or client | One index across all matters | Knowledge management |
| Identity | Scoped keys, rotation, enterprise SSO and RBAC options | Shared associate credentials | IT security |
| Data boundary | Cloud, VPC, on-prem or air-gapped deployment | Privileged text path undocumented | General counsel |
| Legal hold | Configurable retention and technically enforced deletion | No deletion process | Records and risk |
| Audit trail | Request, model and source logging | Outputs not traceable | Risk management |
| Verification | Citations and structured output for review | Unverified AI authorities | Supervising partner |
Frequently asked questions
Does using Plugsky make us compliant?
No. Compliance is your program. Plugsky provides deployable controls - scoped keys, matter separation, private deployment, logging - that you document and audit against your professional obligations.
Can matter files stay inside the firm?
Yes. VPC, on-prem and air-gapped deployments keep prompts, documents and embeddings inside your environment, and the API stays OpenAI-compatible.
How do we enforce conflict walls technically?
Use separate keys and retrieval indexes per matter or client so a workflow cannot query across walls, and review access logs for exceptions.
What should we log?
Request IDs, model names and versions, retrieved source identifiers, outputs and the reviewing lawyer, retained under engagement terms so advice can be reconstructed.
Can AI cite case law reliably?
Citations from retrieved sources are far more reliable than unaided generation, but a lawyer must verify every authority independently before relying on it.
Is firm know-how safe to index?
Yes, in a separate index from matter files, with permissions that match how the know-how is shared today and retention that follows firm policy.
Where should a pilot start?
Pilot on public law and firm know-how with the free plan, verify citation quality and logging, then extend to matter files on private deployment with strict separation.