For legal & general counsel

An improvised legal position is a commitment

When a sales or deal-desk assistant answers a question on liability or an NDA by guessing, the business can act on it before anyone in legal sees it. Keep the approved positions in one place, and every assistant you connect repeats only what counsel signed off.

Repeats only what counsel approved
A versioned record of what was in force, and when
Privileged analysis stays out of scope
Free tier to start

Your AI assistant

Sampleanswering from approved knowledge

Asked

Can we sign a counterparty's NDA with a 5-year confidentiality term?

The approved answer

Our approved fallback is a 3-year confidentiality term for mutual NDAs. Up to 5 years is acceptable if the agreement is mutual and carves out residuals. A one-way NDA in the counterparty's favor goes to counsel before signing.

Grounded in knowledge your team approved

source: legal/nda-fallbacksapproved by General Counsel

Nothing an assistant can repeat until counsel approves it

Draft and in-review positions are never returned to any assistant. A position reaches a connected assistant only after counsel approves it, so it can never surface a stance nobody signed off.

A defensible record of what guidance was in force

Every approved revision keeps its author, reviewer, and approval time. When guidance is disputed later, in a negotiation, an internal question, or litigation, you can show which position was approved and retrievable on a given date instead of reconstructing it from memory.

Scope a token to negotiation-safe guidance

A token can be limited to specific classifications, so a sales-facing or deal-desk assistant reads only approved fallback positions and standard clause language, while privileged or internal analysis stays out of what that assistant can ever retrieve.

Correct one position, every assistant follows

When a fallback position changes, update and re-approve it once. Every assistant calling MCP or REST reads the new approved revision on its next request, with no cached copy or old deck to chase down.

Questions

Should privileged or confidential material go in a system like this at all?: That is counsel's call to make, document by document, not a judgment the product makes for you. Classification controls scope what a given token can retrieve, but whether a particular position belongs in an assistant-retrievable system in the first place, privileged or not, is a decision you make deliberately before it goes in. The product does not make that call. It enforces the scoping you decide on.
Does this replace outside counsel, or make us liable for what the assistant says?: No. The product does not provide legal advice. It serves back only what an approved reviewer put in, and it does not generate new legal positions, evaluate a novel fact pattern, or substitute for a lawyer's judgment on anything outside the pre-approved playbook. The guarantee is narrower and more defensible: an assistant can only repeat what counsel already decided.
Can this plug into our CLM or e-signature tool?: Not today. There is no built-in CLM or e-signature integration. What is shipped is the hosted MCP server and the REST API, and any assistant your team already runs can call either directly. It is a complement to a CLM, not a replacement for one: a CLM manages the contracts themselves, while this serves the approved position that informs them.
What does this cost?: Free is $0 and covers up to 200 documents and 5 members, enough to pilot one playbook end to end, say your NDA fallback positions. Team is $99 a month and raises that to 5,000 documents and 25 members, sized for a legal team maintaining a real playbook across several practice areas. Enterprise is custom priced for higher limits by agreement.
How is this different from a search tool over our playbook folder?: A search index retrieves what is textually similar, including a position that was superseded last quarter, and it has no concept of draft versus approved. This serves only the current approved revision, so an assistant cannot surface a stale or unsigned position alongside the current one.
What does our security review actually find here?: The shipped controls are workspace isolation, classification-scoped tokens, and a full approval trail with author, reviewer, and approval time, all enforced by the retrieval service itself rather than by a prompt. OIDC single sign-on is available on Enterprise. SAML and SCIM are on the roadmap, and there are no published security certifications. A DPA and a formal security review are commitments we can make case by case, not features already in the product. We would rather tell you that plainly than let it be assumed.
Can we use this to approve another team's AI project instead of blocking it?: That is the intended use. When a support or internal-ops assistant might touch a legal-adjacent question, you can point it at an approved, classification-scoped source and know exactly what it is able to repeat, rather than signing off on an unbounded risk that it might say anything. It gives legal a defensible way to say yes.

Planning your first legal playbook pilot?

Tell us which positions you would start with, say NDA fallbacks or liability caps, and who owns approvals, and we will help you scope the pilot and the security review your team runs.

Give every assistant the position counsel approved

Start free, write down one playbook like your NDA fallback positions, approve it, and connect the assistant that needs it.