Three layers of write protection. Read-only by design.

The all-seeing eye for your MongoDB data.

Paste a connection string. Argus samples the schema, generates a draft dashboard, and lets you chat with your data — all without ever writing to your cluster.

How it works

1

Vercel — Next.js frontend

Tambo React SDK drives the generative UI. 7 Zod-validated cards.

2

Cloud Run — self-hosted Tambo

LLM: Gemini 3 Flash. Per-tenant mcp_manager + result_set_guard.

3

Your MongoDB

Read-only user. No writes can reach the cluster.

3-minute demo video

(embed here — 0:45 is the “drop the users collection” moment)

Five insight modules

Each module is generated from the schema, then run as a MongoDB aggregation. Ask in chat, or accept the planner's draft.

Funnel

Conversion rates between top events, computed in a single $facet round-trip.

Cohort

Weekly retention by acquisition week. No SQL, no Tableau, no warehouse.

RFM

Recency-Frequency-Monetary segmentation on a 5×5 grid.

Attribution

First-touch attribution. Markov and Shapley are on the roadmap.

Anomaly

Z-score on daily metrics vs the trailing 28-day mean.

From paste to first dashboard in 2–3 minutes

  1. 1

    Connect

    Paste a connection string. We probe the cluster for reachability.

  2. 2

    Sample

    We sample the top 5 collections and feed the schema to the planner LLM.

  3. 3

    Insight

    You get a draft dashboard. Drag, drop, resize, or ask in chat for more.

Read-only by default. Three layers.

  • 1MCP server runs with MDB_MCP_READ_ONLY=true — write tools are stripped at startup.
  • 2Onboarding recommends a read-only database user.
  • 3argus-result-set-guard blocks $out / $merge at the planner layer, before MongoDB sees them.