north-brook/vesai

VES AI Skill

Use VES AI as a local-first CLI for AI-ready product analytics.

Core Objective

Make product analytics actionable for AI agents by:

1. Discovering high-signal replay sessions.
2. Running replay analysis for session/user/group/query scopes.
3. Pulling supporting evidence from events, schema, insights, errors, and logs.
4. Producing machine-readable outputs (JSON default) and durable workspace artifacts.

Fast Environment Check

Run before any investigation:

vesai doctor
vesai config validate

If config is missing:

vesai quickstart

Quickstart prerequisites:

gcloud auth login
gcloud auth application-default login
gcloud config set project <project-id>

PostHog key requirements:

• User API key from https://app.posthog.com/settings/user-api-keys
• Scope: All access + MCP server scope

Command Map

Replay intelligence:

vesai replays session <session-id>
vesai replays user <email>
vesai replays group <group-id>
vesai replays query [text] [filters]
vesai replays list [text] [filters]

Analytics intelligence:

vesai events
vesai properties
vesai schema data
vesai insights hogql
vesai insights sql
vesai errors list
vesai errors details
vesai logs query
vesai logs attributes
vesai logs values

Config and runtime:

vesai config show
vesai config set <path> <value>
vesai daemon start|watch|status|stop

High-Signal Investigation Workflows

Workflow 1: Replay-first triage

vesai replays list "checkout" --url /checkout --min-active 30 --limit 25
vesai replays query "checkout" --url /checkout --min-active 30 --dry-run
vesai replays query "checkout" --url /checkout --min-active 30

Workflow 2: Deep user story

vesai replays list --email [email protected] --limit 20
vesai replays user [email protected]

vesai replays user contract:

1. Resolve all user sessions.
2. Render each session.
3. Analyze each session.
4. Run one aggregate user inference using all sessions + metadata.

Workflow 3: Group diagnosis + supporting analytics

vesai replays group acme
vesai insights sql "SELECT event, count() AS c FROM events GROUP BY event ORDER BY c DESC LIMIT 20"
vesai errors list --status active

Replay Query Strategy

vesai replays query "<text>" is literal metadata text matching, not semantic understanding.

Use structured filters to make intent explicit:

• --url /checkout
• --email [email protected]
• --group acme --group-key organization
• --where plan=enterprise
• --from <iso> --to <iso>
• --min-active 30
• --limit 50

Always run --dry-run before expensive replay runs when scope is unclear.

HogQL Power-User Guide

How to work

1. Use vesai insights hogql "<question>" to get a generated starting query.
2. Refine with vesai insights sql "<query>" for deterministic iteration.
3. Use vesai schema data --kind events and vesai properties ... to verify field names before complex queries.
4. Keep queries paginated. PostHog MCP SQL results are capped at 100 rows.

Reliable query patterns

Top events:

vesai insights sql "SELECT event, count() AS c FROM events GROUP BY event ORDER BY c DESC LIMIT 20"

Daily activity trend (last 7 days):

vesai insights sql "SELECT toDate(timestamp) AS day, count() AS c FROM events WHERE timestamp >= now() - INTERVAL 7 DAY GROUP BY day ORDER BY day ASC LIMIT 100"

Top pageview URLs:

vesai insights sql "SELECT properties.\$current_url AS url, count() AS c FROM events WHERE event = '\$pageview' GROUP BY url ORDER BY c DESC LIMIT 20"

Enterprise segment example:

vesai insights sql "SELECT event, count() AS c FROM events WHERE person.properties.plan = 'enterprise' GROUP BY event ORDER BY c DESC LIMIT 20"

Key HogQL context for agents

• Base event table is typically events.
• Common columns: event, timestamp, distinct_id.
• Event properties are commonly referenced via properties.<property_key> (example: properties.$current_url).
• Person properties are commonly referenced via person.properties.<property_key> (example: person.properties.email).
• When passing queries in double quotes, escape $ as \$ in shell strings.
• Use LIMIT and OFFSET to paginate if you need more than 100 rows.
• Use --raw on insights sql and insights hogql when debugging raw PostHog payloads.

Agent Output Rules

• JSON is default for data commands. Use --no-json for human-readable output.
• Treat non-JSON output as operator-facing summaries.
• Log exact command + output for each conclusion.
• If output is empty, test adjacent scopes (wider time range, lower filters, remove domain filter) before concluding no signal.

Artifacts and Paths

• Config: ~/.vesai/vesai.json
• Workspace markdown:
  • ~/.vesai/workspace/sessions/
  • ~/.vesai/workspace/users/
  • ~/.vesai/workspace/groups/
• Runtime:
  • ~/.vesai/cache/
  • ~/.vesai/logs/
  • ~/.vesai/tmp/

Troubleshooting

gcloud hash errors (blake2b, blake2s):

export CLOUDSDK_PYTHON=/usr/bin/python3

storage.objects.create permission denied (ADC mismatch):

gcloud auth application-default revoke
gcloud auth application-default login
gcloud auth application-default set-quota-project <project-id>

Missing Playwright executable:

bunx playwright install chromium

Vertex model access errors:

• Verify Vertex AI API is enabled.
• Verify model availability in selected region.
• Confirm project permissions and configured model in vesai.json (gemini-3-pro-preview).

If a schema/analytics tool returns validation errors, fall back to:

1. events and properties for field discovery.
2. insights sql for direct controlled queries.