How SelectStar compares to the tools you already know
An honest, side-by-side comparison of SelectStar against ReAct-loop agents, BI tools, Jupyter + pandas, and Prisma Studio. We're not here to dunk on anyone — these are all great tools. We just do a different job.
Information accurate as of 2026. Submit a PR if anything here is out of date.
The right tool depends on the job
SelectStar isn't a replacement for the tools below — it does a different thing. Here's when to reach for each.
SelectStar
You want an agent that decides what to query, how to profile, what to chart, and how to summarize — with safety guarantees and pure-TS ML.
ReAct-loop agents
You need a flexible general-purpose agent without safety constraints, and you're OK with probabilistic behavior and unbounded loops.
BI tools (Metabase, etc.)
You need recurring dashboards and reports that humans configure once and view many times. Pair with SelectStar for ad-hoc questions.
Jupyter + pandas
You need deep, custom analysis with full Python — scikit-learn, statsmodels, custom transformers. SelectStar's ML is intentionally scoped to quick triage.
Prisma Studio
You just need to browse rows in a Prisma project. Lightweight, no setup. Add SelectStar for the agent layer.
All of them
Most teams use SelectStar alongside a BI tool and a notebook — SelectStar for ad-hoc questions, BI for dashboards, notebooks for deep work. They're complementary, not competitive.
Why a graph beats a loop for databases
Most AI agents run in a ReAct loop: the LLM repeatedly outputs Thought → Tool Call → Observation. That's flexible but unreliable for databases. SelectStar uses a graph with conditional edges.
ReAct loop
- ·LLM gets a list of tools and decides what to call
- ·Can wander, hallucinate queries, or enter infinite loops
- ·Write safety is probabilistic — relies on system prompts
- ·Sequential execution — no fork-join parallelism
- ·Full history passed every step — high token cost
SelectStar graph
- Router classifies intent up front — agents only run if needed
- Deterministic edges — one retry on failure, then clean error
- Write safety is deterministic — Zen mode hard-codes gates
- Fork-join: EDA, Viz, ML run in parallel after a SELECT
- Per-agent scoped context — ~45% token savings
Coming from another tool? Here's the switch.
SelectStar runs alongside your existing tools. You don't have to remove anything.
# You don't have to remove your BI tool, notebook, or Prisma Studio.
# SelectStar runs alongside — they read the same database.
# 1. Clone & install SelectStar
git clone https://github.com/selectstar-dev/selectstar.git
cd selectstar
bun install
# 2. Seed the demo DB & push the Prisma schema
bun run scripts/seed-demo.ts
bun run db:push
# 3. Start the dev server
bun run dev
# 4. Open http://localhost:3000, type 'demo', click Connect
# Or paste your own postgresql:// connection string.From Metabase
Keep Metabase for dashboards. Add SelectStar for ad-hoc questions that don't deserve a saved chart.
From Jupyter
Keep Jupyter for deep work. Use SelectStar for the 80% of questions that just need a quick SQL + chart + summary.
From Prisma Studio
Prisma Studio stays for row editing. Add SelectStar for analysis — it reuses your Prisma client.
From a ReAct agent
If you're building your own ReAct agent and hitting reliability issues, study SelectStar's orchestrator. The graph pattern is the upgrade.
Try SelectStar alongside your existing tools
Open the live app — no install required. You don't have to rip anything out.