Start with the work
Ask for a job-description draft, a policy explanation, a knowledge answer, or help navigating Sorcrr. SAI handles specialist routing behind the interface.
SAI / Sorcrr assistant
One assistant.
Right context.
SAI helps you move through Hiring, Market, Books, and company work without making you learn the machinery behind them. Your request is routed through the permissions and context of the place you are already working.
SAI is an AI assistant. Important outputs require human review; material actions require authorization and confirmation.
Permissioned route
human in control
Your request
“Explain this CRS ordering.”
Authorized context
Current job + viewer-owned policy
Deterministic tool
Fetch ranking receipt; no model required
Explanation for review
Original CRS + lens + reason
The operating idea
Ask naturally. Route carefully.
SAI is one interface, not one unlimited data pool. The platform resolves intent, checks the user’s scope, chooses the narrowest suitable capability, and returns something a person can inspect.
Context is earned, not assumed
SAI uses the current user, company, job, conversation, customer, or provider scope only when that workflow is authorized to use it.
Preview before consequence
Drafts and recommendations remain assistive. Important external, financial, account, or hiring actions keep their normal permissions and confirmation gates.
The request path
Five checks between asking and acting.
The exact route varies by feature, but the trust model stays consistent: establish intent and authority before retrieving context, spending AI credits, or changing data.
Understand intent
SAI separates explanation, drafting, retrieval, navigation, and requested action.
Check access
User identity, role, ownership, company, job, and conversation scope are enforced.
Retrieve narrowly
Only relevant authorized profile, knowledge, customer, finance-support, or hiring context is assembled.
Choose the tool
Use a deterministic capability when possible; invoke a model only when generation or interpretation is needed.
Return control
Show the result, source or explanation, cost boundary, and confirmation required for any material next step.
Permissioned context
Same assistant. Different trust boundary.
Choose a context to see the practical separation. The interface may feel continuous; authorization is not.
Your work, your scope.
Personal SAI can help navigate the platform, explain your own data, draft content, and personalize responses from eligible profile context. Where the profile-context setting is available, you can turn that personalization off.
May useYour authorized profile, current workflow, selected files, and conversations you participate in.
Does not becomeA back door into another user’s private messages, employer-only notes, or company knowledge.
Company knowledge, governed.
Company SAI can use approved company knowledge and authorized operating context for support, sales, hiring, or internal work. Membership and role controls still decide who can see or change each source.
May useApproved knowledge entries, company inbox context, relevant jobs, teams, and workflows.
Does not becomeA merger of employees’ personal memory or an override of company, team, job, or document permissions.
Helpful outside, reviewable inside.
Customer-facing SAI can answer from approved company knowledge and the customer context allowed for that conversation. Company users can review, correct, or take over the thread through governed inbox workflows where enabled.
May usePublished or approved answers, relevant offering or booking context, and the active customer thread.
Does not becomeAccess to owner personal memory, unrelated customer threads, employer-only intelligence, or unrestricted company files.
Built-in memory and context management
Context is assembled. Memory is curated.
SAI does not pour your whole account into every request. It separates what is happening now, what may be remembered for you, what an organization has approved, and the indexes used to retrieve only relevant material.
Session context
Your current request, selected files, recent conversation, and active workflow. Ending or deleting a supported conversation clears its live context without granting access to unrelated sources.
Personal memory
User-scoped preferences, diary or timeline summaries, and other eligible memory can support continuity. It remains distinct from company knowledge and another person’s private context.
Approved knowledge
Personal, job, team, or company knowledge is retrieved only when its source and the current user permit it. Company membership alone does not unlock every document.
Retrieval layer
Embeddings, graph links, and compacted summaries help find relevant context. They do not replace the source record or erase the source’s scope and access rules.
Supported AI settings expose a unified memory inventory, source-level drill-down, retention choices, compaction previews, and bounded cleanup previews. Cleanup coverage varies by source, so the confirmation identifies what will be affected and what remains in its owning product surface.
Tool choice and cost
Not every useful answer needs a model.
SAI is an assistant layer over Sorcrr capabilities. Some are deterministic software; others use a generative model. The route determines whether AI credits are needed.
Deterministic route
Rules, retrieval, and calculations
Examples include applying a saved CRS ranking policy, reading a ranking receipt, navigating to a known workflow, or retrieving exact authorized records. These routes do not need model inference or AI credits.
Model-assisted route
Drafting and interpretation
Examples include drafting a job description, summarizing authorized material, or generating a response from approved knowledge. Variable-cost work may require personal, granted, or company-funded AI credits.
What humans retain
Assistance is not authority.
SAI can reduce busywork and make complex systems legible. It cannot transfer legal or professional responsibility away from the people and organizations using it.
Verify important output
Generated text can be incomplete or wrong. Users must check facts, source material, tone, and suitability before relying on or sending it.
Keep consequential judgment
Hiring, rejection, booking, professional, financial, and customer decisions remain with authorized people—not SAI.
Approve material actions
Normal authorization and confirmation requirements remain in place for saves, sends, purchases, payments, account changes, and other material effects.
Questions worth asking
Know when SAI is in the room.
Sorcrr identifies SAI as an AI assistant. Availability, enabled tools, model providers, and credit requirements can vary by product, account, and workflow.
Is every action inside SAI sent to a generative model?
No. Deterministic tools can answer or act without model inference. Model-assisted routes are used for tasks such as drafting, summarizing, or interpreting authorized context.
Can SAI see everything in my company?
No. Company membership does not erase document, team, job, customer, or role boundaries. SAI receives only the context authorized for the current user and workflow.
Does SAI use my profile automatically?
Eligible profile context may personalize responses in supported personal workflows. Where the user setting is available, you can disable profile-context personalization. Required account operations and deterministic tools remain separate from that preference.
Can I see or remove what SAI remembers?
Supported AI settings show memory inventory and source breakdowns, provide retention controls, and preview eligible compaction or cleanup before a destructive action. Some records are managed in their source surface, and legal, safety, billing, or dispute retention can still apply. The confirmation should identify both affected and preserved sources.
What about interview assistance?
Interview intelligence is a separately gated capability, not a promise that every interview uses AI. Where explicitly enabled, authorized interviewer-only output stays in Internal Remarks; candidates do not receive employer-only intelligence. Humans remain responsible for the interview and decision.
How does SAI relate to CRS?
CRS and Open Algorithm are deterministic. SAI can retrieve and explain their receipts or help configure a policy, but it does not secretly replace the canonical score with a model-generated ranking.
One interface, bounded authority
Let the assistant route the work—not own the decision.
Ask naturally, inspect the result, and stay in control of what happens next.