Assistant

An Assistant that already knows your company

Ask it anything. It reads your docs, sees your systems, respects your access rules, and takes the actions you approved — grounded in your knowledge, grounded in your context.

Live now
Capability 01

Your whole company in the context window

Connected knowledge from Knowledge Studio + a live entity picture from Context Graph + connected apps through Playbooks. The Assistant sees the same operational picture a senior team-mate would: who is asking, what they own, what your policy says, and what write actions are on the table.
grounded answer · sources
answer

Priya’s Okta lockout can be cleared via self-serve MFA challenge — no IT intervention required. If MFA fails, escalate to IT on-call.

cited from
01
IT · Okta account lockout policy
runbook · updated 4d ago
02
HR · Emergency access procedure
policy · updated 2w ago
03
Ticket #48211 · priya · resolved
past ticket · similar match
Capability 02

Speaks your policy, not the internet

Answers come from your sources. The source paragraph is shown alongside the response, one click to open. When the corpus does not contain an answer, the Assistant says so and escalates rather than guessing.
approvals · pending 4
unlock okta account
priya.sharma · self-serve · read-only
denyapprove
refund $128.40
stripe · ord_A4b2c · manager approval
denyapprove
revoke prod IAM role
aws · role/analyst · step-up MFA
denyapprove
reset zoom SSO
kiran.mehta · self-serve
denyapprove
Capability 03

Takes action, with approval

From “reset my password” to “refund this order” in the same conversation. Every action ties to a Playbook; every write is gated by your approval rules; every executed step is logged and reversible.
context · priya.sharma
priyateamon-calloktagithubworkday
Capability 04

Where you already work

Slack, Microsoft Teams, Discord, Zendesk, Intercom, the embeddable widget, and MCP for Claude Desktop / Cursor / any MCP client. One Assistant, seven channels, one consistent voice — including the Colleague persona if you've configured one.
reports · this week
Auto-resolved
78.4%
+4.1 pt
Avg. loop time
11.4s
-2.1s
Approvals pending
12
-3
Coverage
94%
+1.2 pt
monsun
Capability 05

Your model, your key

Choose Anthropic, OpenAI, Google, or a self-hosted model per workspace. Models Studio holds the credential and the guardrails, and rotates models without changing the Assistant's behaviour.
grounded answer · sources
answer

Priya’s Okta lockout can be cleared via self-serve MFA challenge — no IT intervention required. If MFA fails, escalate to IT on-call.

cited from
01
IT · Okta account lockout policy
runbook · updated 4d ago
02
HR · Emergency access procedure
policy · updated 2w ago
03
Ticket #48211 · priya · resolved
past ticket · similar match
Capability 06

Multi-turn memory, per thread

The Assistant remembers what you asked earlier in the same thread — the customer name, the incident id, the ticket you were working on — without leaking context across threads or across users. TTL and per-user scope are configurable.
approvals · pending 4
unlock okta account
priya.sharma · self-serve · read-only
denyapprove
refund $128.40
stripe · ord_A4b2c · manager approval
denyapprove
revoke prod IAM role
aws · role/analyst · step-up MFA
denyapprove
reset zoom SSO
kiran.mehta · self-serve
denyapprove
The BeforeQuery method

Visibility. Reasoning. Action.

Every playbook, every action, every answer follows the same three-beat rhythm — one that keeps every write behind evidence and every decision behind policy.

01

Visibility

Read the full context — who is asking, what they own, what they can access, what breaks if we touch it. Nothing acts on incomplete signal.

02

Reasoning

Match the request to a grounded answer or a playbook. Cite the source paragraph, weigh the risk tier, and route to the correct approver where policy demands it.

03

Action

Execute the write on your systems with an audit-log id, an entity list, and a rollback path. Confirm the outcome with the requester in the same thread.

How every request flows

From ask to resolution in one loop

Six stages every request travels. What changes is how many stages policy lets the agent execute without waiting on a human.

01
Request

In Slack, email, widget, MCP, or the helpdesk. Same voice, same context.

Slack · Teams · Zendesk · Widget · MCP
02
Retrieve

Grounded lookup against your knowledge with source paragraphs held aside.

Knowledge Studio
03
Reason

Read the Context Graph — who, what they own, what breaks if we touch it.

Context Graph
04
Approve

Route to the correct approver per policy. Reversible writes gated, destructive step-up.

Approval matrix
05
Execute

Act on Okta, Stripe, Workday, GitHub, or your own systems. Logged and attributable.

Actions runtime
06
Verify

Confirm the outcome with the requester. Feed the result back into future decisions.

Feedback loop
Enterprise-grade by default

The controls your security team is going to ask about

SOC 2 Type II
audited annually
GDPR
EU data residency
HIPAA-ready
BAA available
SSO / SAML
Okta · Entra · Google
SCIM 2.0
auto-provision
Audit log
every write, traceable
Bring your key
AES-256-GCM at rest
Bring your model
Anthropic · OpenAI · self-host
Zero training
your data stays yours
Rollback
every write reversible
TLS 1.3
in transit
99.9% uptime
SLA on Enterprise

Frequently asked questions

Common questions about BeforeQuery Assistant

It reads your identity from the channel you're chatting on (SSO in Slack / Teams, cookie in the widget, MCP token in an IDE) and applies your workspace's access rules. Restricted documents and cross-user data are filtered before retrieval; the model never sees what you can't.
Yes. Pick a named AI Employee (Remy, Ivy, Harry, Fin, Sig, Lex, Sol, Mel, Rev, Dex) and the Assistant adopts that voice, that avatar in Slack, and that signature. Or write a custom system prompt in Agent Studio.
It says so. The abstention gate refuses to guess when retrieval verdicts come back below threshold. You can configure the fallback: escalate to a human queue, ask a clarifying question, or route to a specialist Colleague.
No. Your data stays in your workspace. Provider calls run under zero-retention API terms; on Enterprise, BYOK routes traffic through your own account under your own agreement.
Yes. Every knowledge base ships an MCP endpoint. Add it to Claude Desktop or Cursor and your Assistant is a first-class tool alongside your local files and the model's own capabilities.

Run assistant on BeforeQuery

Book a demo and see what BeforeQuery Assistant does on your own data — usually within 45 days.