Docs explain your product. OneTwoAgent also recognizes the signed-in customer, uses the account data you authorize, and makes scoped fresh backend lookups when the answer depends on what is happening right now.
14-day free trial · 50 AI conversations · no cardRead the signed-in customer docs
Why did my latest import fail?
Docs-only AI
Imports usually fail because of unsupported date formats, missing columns or files over the size limit. See “Troubleshooting CSV imports”.
OneTwoAgent
Your latest CSV import (imp_8421) failed: some dates aren’t in a supported format. Export them as YYYY-MM-DD and run the import again.
Account-aware AI customer support answers with two kinds of evidence: your product knowledge, and facts about the signed-in customer’s own account — plan, subscription, access, usage, jobs and integrations — that your backend authorizes.
OneTwoAgent receives those facts in a signed account snapshot when the customer is identified, and, when an answer depends on current state, from a scoped fresh lookup to one read-only endpoint in your backend. When the evidence or the authority isn’t there, it hands the conversation to your team.
Docs tell your AI how the product works.
Account context tells it what is happening to this customer.
The same three questions, answered from docs alone and with authorized account evidence.
Docs-only answer
The Salesforce integration is available on eligible plans. See “Integrations by plan”.
Account-aware answer
Salesforce sync isn’t part of your Pro plan, so it isn’t available on your workspace. “Integrations by plan” lists the plans that include it.
Docs-only answer
Common causes are unsupported date formats, missing columns and files over the size limit.
Account-aware answer
Your latest CSV import (imp_8421) failed with INVALID_DATE_FORMAT: some dates aren’t in a supported format. Export them as YYYY-MM-DD and run it again.
Docs-only answer
You can see your usage under Settings → Billing → Usage.
Account-aware answer
You’ve used 42 of your 50 imports, so 8 are left.
Example workspace (Tallyworks). Account facts follow the saas_support_v1 contract.
Five account sections, each a small, typed projection. Your backend decides which sections to enable and which facts to fill — the agent only receives the data you authorize.
Which plan applies and whether the subscription is trialing, active, past due, cancelled or expired.
section: subscription
The customer’s role and the capabilities their account can use.
section: access
Current usage against the limits you choose to show.
section: usage
Recent imports, exports, syncs or other jobs: pending, running, completed, failed or cancelled.
section: jobs
Whether each connection is connected, disconnected, pending or in error.
section: integrations
These are the sections the contract supports, not data every integration exposes automatically. A section you don’t enable is unknown to the agent — not zero, and not “no access”.
Your server sends a small set of facts when it identifies the customer. OneTwoAgent validates them and signs them into the identity token. Most account questions are answered from it, with no extra call.
When an answer depends on current state, the agent asks your backend for one allowed section — and gets an up-to-date read.
Fresh fields replace that snapshot section for the current turn. If the check fails or times out, the agent says it couldn’t check the current state — older facts are never presented as a live result.
Your server vouches for who is asking. OneTwoAgent combines product knowledge with the account evidence you authorize — and only answers what that evidence supports.
Signed-in customer
Alex Rivera in your app
Your server
Identity exchange: user, workspace and the account facts you choose
OneTwoAgent support agent
One conversation per user and workspace
Evidence for this turn
Product knowledge
Help Center, website, Google Drive, files
Authorized account snapshot
plan · usage · jobs · integrations · ≤ 5 min
Scoped fresh backend lookup
get_account_status({ section })
Human handoff
When evidence or authority is missing, your team takes over in the inbox.
OneTwoAgent never connects to your database. Your application exposes the specific support context and scoped reads the agent may use — and nothing else.
Your app
Authorized support contract
OneTwoAgent
Receives only enabled sections
Integrations isn’t enabled here, so the agent can’t see it — those questions go to your team.
You pick the sections and project each field yourself. The agent can only ask for a section you enabled.
Your endpoint verifies the live session and workspace membership on every lookup. Sign out or remove a member and the next lookup fails.
No SQL, no write actions, no full records, no hidden flags or stack traces — only the closed schema your endpoint returns.
The identity secret and a separate read credential live in your server environment. The browser only gets a short-lived token signed by OneTwoAgent.
Your endpoint answers one section (abridged)
{
"version": 1,
"section": "jobs",
"subject": {
"userId": "usr_5d1c90",
"workspaceId": "ws_ridgeline"
},
"…": "operation, turn and grant bindings"
}{
"section": "jobs",
"data": {
"jobs": [
{
"id": "imp_8421",
"type": "csv_import",
"state": "failed",
"reasonCode": "INVALID_DATE_FORMAT"
}
]
}
}What happens when a signed-in customer asks about something only their account can explain.
Example workspace. The thread uses the widget’s own message styles; on a handoff the agent sends no scripted hand-over message.
Account-aware doesn’t mean the AI answers everything. It answers when the evidence and the authority support it — and brings in your team when they don’t.
The question needs a section you haven’t enabled. The agent doesn’t guess.
Refunds, plan changes, approvals. Account lookups are read-only — the agent can’t take those actions.
A lookup failed, timed out or the proof expired. Older facts are never passed off as current.
When the customer asks for a human, the agent hands over instead of carrying on.
Most teams start with knowledge and identity, then add account context and fresh lookups for the questions that need them.
The CLI also installs a coding-agent skill. It inspects your app and adapts the integration to the auth system you already use.
npx -y @onetwoagent/cli@latest init --browseronetwoagent account-lookup configure descriptor.json“Is this included in my plan?”
subscription
“How much usage do I have left?”
usage
“Why can’t I open this feature?”
access
“Why did my latest import fail?”
jobs
“Is my integration connected?”
integrations
What should happen.
Docs, Help Center, website, files
Who is asking and what applies to them.
Signed identity + signed snapshot
What is happening right now.
One scoped, read-only section
What to do when AI shouldn’t resolve it alone.
Shared inbox, full context
The full picture: AI customer support for SaaS. Weighing an existing tool? See OneTwoAgent vs Intercom, the Fin AI alternative and pricing — fresh lookups don’t add a separate charge.
AI customer support that answers with your product knowledge and with facts about the signed-in customer’s own account — such as plan, access, usage, jobs and integrations — that your backend authorizes. It can explain what is happening for this customer, not only how the product works.
Yes — in the OneTwoAgent website widget. Your server exchanges its own session for a widget identity token that lasts 15 minutes, and the browser calls identify(token). Each conversation is bound to one user and one workspace; the identity secret never reaches the browser.
Yes. It uses the account facts your backend authorizes — from a signed account snapshot and, for current state, from scoped fresh lookups to your backend. The supported sections are subscription, access, usage, jobs, integrations.
No. It never connects to your database. Your backend sends a small snapshot of the facts you choose and, optionally, answers read-only lookups for one allowed section at a time from your own endpoint, using your existing permission checks.
When an answer depends on current state, the agent requests one allowed section from a single read-only HTTPS endpoint in your backend. Your endpoint rechecks the live session and workspace membership, then returns a closed schema. Results are valid for at most 30 seconds, and there is at most one lookup per customer turn.
Account context is a signed snapshot your server provides when it identifies the customer; it expires within 5 minutes and answers most questions with no extra call. A fresh lookup is an on-demand read of one section when the answer depends on what is happening right now. Fresh results replace that snapshot section for the current turn.
Only what you expose: plan and subscription status, account status, role and capabilities, usage metrics with limits, recent jobs with their state, and integrations with their state — each with an optional reason code. Anything you leave out is unknown to the agent. Credentials, tokens, stack traces and other users’ data must never be included.
No. Account proof stays in the widget’s memory, not in local storage or URLs. Long-term AI summaries, shared caches and raw replay captures are disabled for account turns, and earlier messages are never treated as current account facts. Your inbox keeps the conversation history.
Yes. When a section isn’t exposed, the current state can’t be confirmed, the request needs human authority or the customer asks for a person, the agent stops instead of guessing. The conversation is marked as needing a reply in your shared inbox, your team is notified, and a teammate takes over with the full context.
Yes. You keep your auth and data model. The CLI and its coding-agent skill adapt the integration to your existing server session; you add the snapshot to your identity endpoint and, optionally, one read-only lookup endpoint.
Start with your product knowledge. Connect signed-in customer context when you’re ready. 14-day free trial · 50 AI conversations · no card.