FAQs

An FAQ is a named list of question-and-answer pairs you wrote. Attach it to an agent or an agent task with config.faqs: the questions go into the prompt, and the agent fetches the written answer when one is asked. Use it for stable facts asked often — hours, refund policy, delivery areas. Behaviour belongs in the prompt; per-caller data belongs in a tool. There is no document upload or retrieval over files.

Create an FAQ

Dashboard: Build → FAQs → New FAQ. API:

curl -X POST https://api.talqing.com/v1/faqs \
  -H "Authorization: Bearer $TALQING_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "name": "Store FAQ",
    "entries": [
      { "question": "What are your opening hours?", "answer": "We are open nine to six, Monday to Saturday." },
      { "question": "Do you deliver on Sundays?", "answer": "No. Orders placed on Saturday go out on Monday." }
    ]
  }'

const faq = await talqing.faqs.create({
  faqDefinition: {
    name: "Store FAQ",
    entries: [
      { question: "What are your opening hours?", answer: "We are open nine to six, Monday to Saturday." },
      { question: "Do you deliver on Sundays?", answer: "No. Orders placed on Saturday go out on Monday." },
    ],
  },
});

faq = talqing.faqs.create(
    name="Store FAQ",
    entries=[
        {"question": "What are your opening hours?", "answer": "We are open nine to six, Monday to Saturday."},
        {"question": "Do you deliver on Sundays?", "answer": "No. Orders placed on Saturday go out on Monday."},
    ],
)

The response is the FAQ (id, name, entry_count, …) with entries, each with its own id. Writing entries: phrase questions the way a caller asks them (paraphrases need no extra entries); keep voice answers to one or two speakable sentences; do not repeat the answers in the prompt.

Editing entries

Entries are appended or edited individually, never replaced as a list:

# Append (1–500 entries, all or nothing)
curl -X POST https://api.talqing.com/v1/faqs/7d1c9e52-3b4a-4f86-a0d7-5e8c2b6f9a13/entries \
  -H "Authorization: Bearer $TALQING_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"entries": [{"question": "Is there parking?", "answer": "Yes, free parking behind the store."}]}'

# Change one entry
curl -X PATCH https://api.talqing.com/v1/faqs/7d1c9e52-3b4a-4f86-a0d7-5e8c2b6f9a13/entries/c41e7a90-6d2b-4b35-8f1a-9e0d3c5b7a24 \
  -H "Authorization: Bearer $TALQING_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"answer": "We are open nine to seven, Monday to Saturday."}'

The dashboard’s Import CSV appends a file with header question,answer (up to 500 rows) as one all-or-nothing append. The API has no import route; send the rows to POST /v1/faqs/{faq_id}/entries.

Attach it

{ "faqs": [{ "faq_id": "7d1c9e52-3b4a-4f86-a0d7-5e8c2b6f9a13" }] }

faqs is a list, so a PATCH replaces it whole. Attaching is a config change: publish the agent for callers to get it. An item may carry an inline faq (name plus at least one entry) instead of faq_id:

Sent on	Result
Agent or task create/update	Created as a stored FAQ; the saved config holds its `faq_id`. A taken name is a `409`
A call request (`agent`, `agent_override`, `agent_team`, batch, chat)	Used for that session only. On a batch, prefer `faq_id` — the plan is stored per call

How the agent uses it

At session start, the questions of every attached FAQ are appended to the system prompt under ## FAQ with short ids (q1, q2, …). The agent gets a get_faq_answers tool ({"ids": ["q3"]}) returning the written answers.

The questions cost tokens every turn; answers only when fetched.
On voice and video the agent says it is checking in the same turn as the tool call.
FAQ text is never templated — {{userdata.name}} in an answer is read literally.
A question matching nothing is answered from the prompt and tools as usual; write a refusal rule into the prompt if you want one.
get_faq_answers is a reserved tool name.

Edits are live

FAQs have no draft or versions. Entry changes reach every attached agent and task from its next session, without publishing; a running session keeps what it loaded. Publishing an agent freezes which FAQs are attached, not their content.

A wrong answer saved to an FAQ is what the next caller hears. Reading is open to every member; writing needs the editor role.

Deleting

DELETE /v1/faqs/{faq_id} is refused with a 409 while any agent’s or task’s draft or current published version attaches it (this FAQ is attached to agent '…' — detach it there before deleting it). Older versions do not block it.

Limits and errors

Limit	Value
`name`	1–80 characters, unique in the workspace (case-insensitive)
`question`	1–300 characters, unique within the FAQ (case-insensitive, trimmed)
`answer`	1–2,000 characters
Entries per FAQ / per append	500 / 1–500
FAQs per agent or task	10

Cause	Status
Name already used	`409` `an FAQ named '…' already exists`
Duplicate question in the request, or one the FAQ already has	`422`
Append past 500 entries	`422`
Unknown `faq_id`, or the same FAQ attached twice	`400`
Attached FAQ with no entries	Warning at validate/publish

Endpoints: GET/POST /v1/faqs, GET/PATCH/DELETE /v1/faqs/{faq_id} (PATCH renames), POST /v1/faqs/{faq_id}/entries, PATCH/DELETE /v1/faqs/{faq_id}/entries/{entry_id} — SDK faqs.* and faqs.entries.*. See the API reference.

Prompting

What belongs in the prompt instead.

Versions

What a publish freezes.

Get started

Build

Guides

Create an FAQ

Editing entries

Attach it

How the agent uses it

Edits are live

Deleting

Limits and errors

Next

Prompting

Versions

​Create an FAQ

​Editing entries

​Attach it

​How the agent uses it

​Edits are live

​Deleting

​Limits and errors

​Next

Prompting

Versions

Create an FAQ

Editing entries

Attach it

How the agent uses it

Edits are live

Deleting

Limits and errors

Next