Reports from Multiple Sources
Integrate your data warehouse, product analytics, billing, search analytics, and observability into one swarm — then ask it the questions your team would have asked a BI tool. Charts render as auto-hosted Pages.
The swarm joins data across ClickHouse, Postgres, Convex, Clerk, PostHog, Stripe, Google Search Console, GitHub, and SigNoz, and answers scheduled questions ("daily product-growth update?", "weekly GTM metrics?") and on-demand questions in Slack ("how many active orgs signed up last week?").
What it does
Reports render as auto-hosted HTML Pages with charts when complex, or Slack threads with chart replies for recurring digests.
Agents
- Lead — orchestrates; holds the scoped secrets needed for cross-source queries.
- Researcher — synthesis: takes raw data + the question, produces the narrative.
Tools & Skills
Built-in (ships with agent-swarm)
- agent-fs (archives), swarm KV (question cache + report state), Pages (auto-hosted HTML reports with charts),
script-run+swarm-scriptnodes (deterministic catalog scripts — JSON stdout merges into node output for downstream interpolation). Use these for "fetch + JSON" sections so the LLM stays out of the loop where it adds no value.
Custom (swarm-managed)
- Integration skills (one per source):
gsc-analytics(Search Console),posthog-interaction(PostHog HogQL — also joins Convex / Clerk / Stripe warehouses),clerk-analytics(user/org growth from Clerk),turso-interaction(Turso LibSQL state),signoz-interaction(SigNoz observability),granola-api(Granola meeting signal),telemetry-report(charts from internal telemetry). daily-growth-snapshotswarm-script — calls every source over HTTP/SQL and emits a unified JSON payload, so the agent-side report is mostly framing, not data-pulling. Date-window discipline (align to complete UTC days, exclude today) is encoded in the script.- Chart-generation skill — renders the JSON payload into chart images embedded in the Page or Slack reply. Backed by Chart.js running inside a
swarm-scriptnode so output is deterministic and snapshot-comparable.
Workflows / Schedules
daily-product-growth-update— daily. Joins ClickHouse (CLI installs) + Clerk (signups) + Postgres (product activity) + PostHog → Convex (warehouse) + GitHub (stars/forks/traffic). Posts a Slack thread: header + snapshot + up to 8 chart replies + TL;DR.gtm-weekly-review— weekly. GitHub stats + Plausible + Google Search Console as a research thread.daily-hn-briefing— daily. Browser-automation scrape of Hacker News, filter for AI/dev-tools relevance, email via AgentMail, archive in agent-fs.memory-rater-daily-digest— daily. Internal-tool health digest (relevant to swarms with a memory subsystem).daily-blocker-digest— daily prelude that verifies operational-runbook items.
Patterns used
- No-op When Nothing Changed — digests skip on a quiet day instead of posting an empty report.
- For preparing a supplier renewal briefing from several documents for a human owner, see the supplier renewal decision guide.
The same fetch-then-synthesize split applied to booking records, where a person owns any change, is worked through in Investigating booking discrepancies with AI agents.
Applied to a guest's accessibility question, the same split gathers the guest's words and the approved property facts, and a person verifies each fact before anyone answers. See the guest accessibility request handoff guide.
Tips for new swarm users
- Use deterministic
swarm-scriptnodes for "fetch + JSON" sections. Keep the LLM out of the loop where it adds no value — faster, cheaper, more reliable. - Always exclude the partial current day from time-windowed metrics. Today's numbers are incomplete and the agent will draw wrong conclusions.
- Render multi-chart reports as a Page, not a Slack thread. People want skimmable: a linked HTML page with 8 charts beats 8 image replies.
- One agent owns the cross-source secrets (scoped per agent), keeping the secret surface small and auditable.
- Cache slow warehouse queries in KV with a short TTL — a 1-hour cache is invisible to users and saves a lot of cycles.
- Ask the swarm questions in Slack and let it pick the source — a well-skilled lead routes "active orgs last week?" to Clerk, "activation funnel?" to PostHog, "top GSC query?" to Google Search Console, without you naming the source.
- The same collection plus synthesis pattern can assemble evidence beyond reporting: deterministic steps gather a supplier invoice's records and compare the lines with its purchase order, then an agent writes the pack a finance owner reviews in Reviewing supplier invoice and purchase order discrepancies with AI agents.
- A product catalog is another place this split earns its keep: deterministic steps compare the approved catalog record with what each channel lists, and an agent puts every difference in front of a catalog owner with a source and a read timestamp, as worked through in Reviewing inconsistent product listings across sales channels with AI agents.
- A rights review is another place this pattern earns its keep: deterministic steps gather an asset's license agreement, its asset record and the requested use, then flag a missing amendment version or a channel, territory or period that disagrees, and an agent puts each one in front of a qualified rights owner with a source and a read timestamp, as worked through in Preparing a digital asset license review for a human rights owner with AI agents.
- Point the same collection step at a shift change. For handing unresolved operational work from one shift to the next, deterministic collection feeds an item list with a source link and a named owner, and the incoming lead accepts each item explicitly. See Handing off unresolved operational work between shifts with AI agents.
- The same collection plus synthesis pattern can assemble the evidence for a late delivery: deterministic steps gather a delayed order's records and flag a missed promised date or a missing scan, then an agent writes the handoff an operations owner reviews in Coordinating delivery delay investigations between operations and support with AI agents.
- The same collection plus synthesis pattern can assemble the evidence for a disputed product return: deterministic steps gather the order, the return authorization, the inspection record and the policy version and compare them, then an agent writes the brief an operations owner decides on in Preparing a disputed product return for a human decision with AI agents.
- The same collection plus synthesis pattern can assemble the evidence for a partner payout exception: deterministic steps match a payout's transactions, fees and refunds within a settlement window, then an agent writes the packet finance and partner operations review in Reviewing marketplace partner payout discrepancies with AI agents.
- The same collection plus synthesis pattern can assemble the context for a moderation appeal: deterministic steps gather the listing as decided, the original decision with its reason code and the dated policy in force, then an agent writes the source-linked packet a named reviewer rules on in Preparing marketplace moderation appeals for human review with AI agents.
- Point the same collection plus synthesis pattern at a capacity review. Deterministic steps gather approved demand and capacity records for one planning period, then an agent writes the source-linked brief that operations and engineering decide from in the service capacity planning brief guide.
- The same collection plus synthesis pattern can assemble the context for one property maintenance request. Deterministic steps gather the guest's report, the unit and appliance record, availability, prior work on the asset and the approved procedure and flag what is missing or conflicts, then an agent writes the source-linked packet an operations owner reviews before deciding to dispatch in Preparing a property maintenance request for human triage with AI agents.
- The same collection plus synthesis pattern can assemble the evidence for a disputed charge: deterministic steps gather a disputed charge's invoice, agreement and metered usage and compare the billed quantity against them, then an agent writes the review packet a finance owner and the account owner validate in Reviewing customer billing disputes with AI agents.
- The same collection plus synthesis pattern can assemble the evidence for an unmatched bank transaction: deterministic steps compare a statement line with the ledger entries inside a date window and flag a fee deduction, a partial settlement or a missing record, then an agent writes the packet a finance owner validates in Preparing unmatched bank transactions for finance review with AI agents.
- The same collection can assemble one ticket's escalation packet. Escalating a customer support issue to engineering with AI agents gathers one ticket's history, the documentation for the reported behaviour and matching engineering issue references into a packet that a support owner validates before an engineering owner accepts it.
- The same collection plus synthesis pattern can assemble a supplier's periodic performance record: deterministic steps count deliveries, late arrivals and quality incidents for one period, then an agent writes the source-linked packet an operations owner reviews in Reviewing supplier performance across delivery, quality and service records with AI agents.
- The same collection plus synthesis pattern can assemble the differences on a supplier's order confirmation: deterministic steps match one approved purchase order against the confirmation the supplier returned and flag a short quantity, a moved delivery date, a changed unit price or an order revision that does not agree, then an agent writes the source-linked packet purchasing and operations review in Reviewing purchase order and supplier confirmation differences with AI agents.
- The same collection plus synthesis pattern can assemble the evidence for an ingredient specification change: deterministic steps collect a supplier's new specification and the last approved version, compare them field by field and look up the recipe records that use the ingredient, then an agent writes the source-linked packet a food-safety owner reviews in Reviewing a supplier ingredient specification change with AI agents.
- The same collection plus synthesis pattern can assemble a partner launch readiness check: deterministic steps gather the approved onboarding requirements, listing status, payout setup and open support items and flag missing fields or dates, then an agent writes the source-linked checklist an operations owner decides launch or hold from in Reviewing marketplace partner launch readiness with AI agents.
- The same collection plus synthesis pattern can assemble the evidence for a missed service response window: deterministic steps gather the agreement version in force, the ticket timestamps and the service records for one customer and period and compare them with a configured window, then an agent writes the packet a named reviewer and the account and finance owners decide from in Reviewing a potential customer service-level breach with AI agents.
- The same collection plus synthesis pattern can assemble the evidence for a proposed customer price change: deterministic steps gather the agreement, the approved price list and the billing records and flag inconsistent dates or missing terms, then an agent writes the review packet a finance owner and the account owner decide from in Reviewing a proposed customer pricing change with AI agents.
- The same collection plus synthesis pattern can assemble the evidence for a bid or no-bid decision: deterministic steps gather the RFP requirements, the approved capacity records and the internal capability evidence, then an agent writes the requirement-to-evidence packet a commercial owner decides from in Preparing a bid or no-bid review for an incoming RFP with AI agents.
- The same collection plus synthesis pattern can assemble the evidence for a contractor's credential check: deterministic steps gather the contractor's records, the assignment's requirements and the supplied certificates and flag a missing document, a name that differs between them or an expiry that falls inside the assignment window, then an agent writes the source-linked packet a named operations owner decides assignment or hold from in Preparing contractor credential expiry checks for human review with AI agents.
- The same collection plus synthesis pattern can assemble a freshness review for one partner feed: deterministic steps compare the newest source timestamp with the expected update interval, record the last successful ingestion and flag unknown values, missing records and timezone differences, then an agent writes the source-linked packet engineering and operations decide from in Investigating a stale partner data feed with AI agents.
Related
Related guide: post-incident business impact review applies the same deterministic collection to an outage window, with a source on every line.
- For a worked application of this pattern, see how agents help investigate failed payments. The brief goes to a named operations owner who decides every financial step.
- Point the same collection step at customer records. Turning customer feedback into product decisions applies the same deterministic fetch to support records and call notes across accounts, then has a product owner verify each theme before anything reaches a roadmap.
DORA Metrics
Run recurring DORA metrics reports for any Git repository. The swarm computes exact deployment frequency and lead time from release tags, labels CFR and recovery time as proxy estimates, and updates one stable Page URL in place.
Self-Documenting & Release Reports
Keep your docs fresh automatically, generate release notes from real commits, and produce release videos with Remotion + browser-automation captures. No-op silently on quiet days.