Using picoask

1Getting started

Sign in at picoask.ai. On your first visit you'll set a few preferences — how you like reports written (statistical detail, plain language, or both) and the units you think in — which picoask uses to shape every answer it writes for you.

Every new account includes read-only sample projects — a revenue dataset and a support-operations dataset with a year of realistic history. They're the fastest way to feel how picoask investigates a question before you connect anything of your own: open one, ask "why did revenue dip last month?", and watch it work.

2Connect a data source

picoask connects to PostgreSQL, MySQL, Amazon Redshift, Snowflake, and BigQuery. It always connects with a read-only role — and verifies the role really is read-only before the project can go live.

Create the role in your database and grant it SELECT on the schemas you want analyzed. For PostgreSQL:

CREATE USER picoask_ro WITH PASSWORD '<strong-random>';
GRANT CONNECT ON DATABASE <db> TO picoask_ro;
GRANT USAGE ON SCHEMA public TO picoask_ro;
GRANT SELECT ON ALL TABLES IN SCHEMA public TO picoask_ro;
ALTER DEFAULT PRIVILEGES IN SCHEMA public
  GRANT SELECT ON TABLES TO picoask_ro;

Repeat the USAGE + SELECT grants for any additional schemas. The connect step in the app shows the equivalent snippet for MySQL, Redshift, Snowflake, and BigQuery (where access is granted to a service account via IAM instead of a SQL user), pre-filled with the database and username you type.

If your database restricts inbound connections, allowlist picoask's egress IPs in your firewall or security group. Connections always originate from a small, stable set of published IPs — the current list is shown in the "Before you connect" panel of the connect step.

3Onboard your project

Once connected, picoask learns your data in a guided pass:

• Pick tables. Choose which tables belong in the project (up to 100 per data source). On large databases picoask recommends an initial set matched to what you say the project is for.
• Review descriptions. picoask profiles each table and drafts a plain-English description of every table and column. Approve each one, correct it in a short chat, or reject it — nothing is used until you've seen it.
• Get a metric graph. It proposes your business metrics and how they drive one another, with the evidence for each edge. This graph is what makes root-cause questions land on real relationships.
• Get a dashboard. A starter dashboard is generated from the metric graph — edit it, or ask for new panels in plain language.

You can leave onboarding at any point and resume where you stopped; drafts are kept per project.

4Ask questions

Ask in plain language, as specifically as you like — "why did recurring payment volume drop in June, and is the decline significant?" beats "show me revenue". If the question is ambiguous, picoask asks one clarifying question rather than guessing. Every answer arrives with its evidence: the queries it ran, the tables it built, the statistical tests it applied, and a suggested next step.

The composer supports slash commands:

• /new — start a fresh session.
• /retry — re-run the last question.
• /share — share the session as a public read-only link, or with an org member.
• /memory — save a fact your data can't know (a pricing change, an outage, a campaign launch). It's used as context in future analyses.
• /recall — see what picoask remembers.
• /forget — find a remembered fact and remove it.
• /help — list all commands.

Reports can be exported, shared as links, or turned into scheduled reports that re-run and deliver to your email.

5Manage your project

• Add sources. A project can span multiple data sources; add one without re-onboarding the others, and picoask proposes cross-source joins.
• Scan for changes. picoask watches your schema for drift automatically, and you can run a scan any time: new and altered tables appear as drafts with AI-drafted descriptions you approve, refine in chat, or reject.
• Curate the schema. Exclude tables you don't want analyzed — what's excluded is never described or queried.
• Work as a team. Invite teammates to your organization and grant them access per project.

6Ads & analytics integrations

First-party connectors ingest Meta Ads, Google Ads, and GA4 into a managed warehouse attached to your project, unified into one ad-reporting model with the standard metrics (CPC, CTR, ROAS, CAC) pre-defined. Enable a connector during onboarding or later from the integrations page; picoask backfills history and keeps it synced.

Step-by-step credential guides for each platform:

• Connect Meta Ads — Business Manager system user + long-lived ads_read token.
• Connect Google Ads — developer token, OAuth client + refresh token, Customer ID.
• Connect Google Analytics 4 — service account with Viewer access on the property.

7Billing & usage

Every plan includes a monthly analysis budget; bigger, more complex questions use more of it, and you can see where you stand in settings. Paid plans can enable overage so a big question never stops mid-analysis. Plans and inclusions are on the pricing page; invoices and payment methods live in your workspace settings.

8Security & data access

• Read-only, verified. picoask confirms the connection role cannot write — checked before every query, not just at setup.
• Scoped visibility. Excluded tables are never described, queried, or seen by any agent.
• No LLM sees raw data. Raw rows stay inside the query and computation layer — LLM agents work from your schema, table descriptions, and computed results only.
• Known egress. All database connections originate from picoask's published egress IPs, so you can allowlist precisely.
• No training on your data. Connected data is not used to train foundation models without your explicit consent — see the Privacy Policy.