PromptIQ Studio
Start free

StudioChat and voice agents, from one prompt

Agents that answer from your documents — and pick up the phone

Build a chat agent or a voice agent from one plain-English prompt, point it at the documents you already have, and see the passage behind every answer.

Start with 20 free runsSee pricing
  • Cited answers
  • Chat and voice
  • Your own provider keys

No card. One chat agent and one voice agent on the free plan.

Support agent · billing workspacelive

What's the refund window for an annual plan?

Annual plans can be refunded in full within 30 days of purchase[1]. After that the remaining months are credited rather than returned[2].

grounded · 0.942 sources1.2 sgpt-oss-20b
[1]billing-policy.pdfp. 4
[2]terms-of-service.md§ 7.2

Channels

Two ways people will reach your agent

The same agent and the same documents behind both. Pick one to see the surface it puts in front of somebody.

Onboarding assistant/shared/8f3c…public link · no account

How much notice do I need to give before leaving?

Four weeks for salaried roles, and two weeks during a probation period[1].

[1]employee-handbook.pdfp. 12

And during probation?

Two weeks

Ask anything about the handbook…

The path an answer takes

Checks that run on the answer, not on the log

Screening after the fact tells you what went wrong. These run before the reply reaches anyone.

  1. Question

    Typed in chat, or spoken on a call

  2. Retrieve

    Passages pulled from your own documents

  3. Knowledge base

    Uploads and S3, indexed as they arrive

  4. Check

    PII stripped, toxicity screened, answer scored

  5. Tools

    Systems the agent may reach mid-answer

  6. Answer

    Returned with the sources it used

Observability

See why an agent said what it said

Every turn is recorded with its retrieval trace, latency and token cost — including the ones that went wrong.

Observability / Runs

Conversations

1,284+12.4%

last 7 days

Median latency

1.2sp95 1.4s

chat and voice

Grounded

0.91−0.02

mean score

Credits used

3,120of 5,000

growth plan

Turns per day

totalused a tool
Mon
Tue
Wed
Thu
Fri
Sat
Sun
TimeAgentChannelTokensLatencyGroundedStatus
14:22:08Support agentchat1,8420.9s0.96ok
14:21:44Reception agentvoice2,1101.3s0.93ok
14:20:57Onboarding assistantchat9660.7s0.98ok
14:19:31Support agentchat1,2041.1s0.42ungrounded
14:18:02Billing agentchat1,5302.8s0.89ok
14:17:19Reception agentvoice———failed

In the studio

Everything an agent needs, in one workspace

Knowledge that stays current

Each agent gets its own document workspace. Upload files, or point it at an S3 folder and new objects are indexed on their own — nothing is ever silently removed.

billing-policy.pdf2.1 MBindexed
employee-handbook.pdf4.7 MBindexed
q3-pricing.xlsx318 KBindexing

Chat agents

Describe how an agent should behave in plain English — one prompt, no boxes to fill in. It answers from the documents you attach, with numbered citations.

Voice agents

Real-time calls with server-side speech, so quality does not depend on the caller’s browser. Interrupt it and it stops and listens, the way a person would.

Tools

Attach the systems an agent may reach while it answers — send the message, read the row, open the ticket.

SlackGmailGoogle SheetsGoogle CalendarGitHubPostgreSQLAmazon S3HTTParXiv

Skills

Turn on what an agent does while composing: source citations, analysis over spreadsheet rows, conversation memory, charts rendered inline.

Observability

Every turn is recorded with its retrieval trace, latency and token cost, so you can see why an agent said what it said instead of guessing.

retrieve410 ms
tool · run_sql260 ms
compose520 ms
Request inspectorturn 41c8

As typed

Call me on +44 7700 900142 or email rachel.doyle@example.com about the invoice.

Sent to the model

Call me on [PHONE] or email [EMAIL] about the invoice.

toxicityclear
pii2 entities redacted
groundedness0.94

Responsible AI

Guardrails that leave a record

Four checks on every request, each one reported with the turn rather than buried in a log nobody opens.

PII redaction
Personal data is stripped from prompts and responses before they leave the request.
Toxicity filter
Requests and replies are screened, not just logged after the fact.
Groundedness check
Answers are scored against the retrieved passages, so an ungrounded reply is visible instead of confident.
Your keys, your models
Bring your own provider keys. Nothing is shared between organizations.

Build your first agent in a few minutes

Twenty runs is enough to attach a document, ask it real questions and hear it answer out loud.

Start freeSee pricing

No card. One chat agent and one voice agent on the free plan.