What is OpenDashboards?
OpenDashboards is a Next.js app from Thesys, the company behind the C1 generative UI API and the OpenUI component standard. You connect data sources, type a question such as "how was revenue this month?", and a Build agent explores the tables, runs test queries and asks Thesys' OpenUI Dashboards API to draw a dashboard with filters and drill-downs. Each chart keeps its query, so the dashboard reloads live data every time it opens, and a follow-up message rebuilds only the chart you asked about.
Built-in connectors cover PostgreSQL (including Supabase and Neon), MySQL and MariaDB, ClickHouse, BigQuery, PostHog, Google Analytics 4 and Search Console. Anything else, such as Stripe or HubSpot, comes in through a remote MCP server over HTTPS with API-key authentication; OAuth is not supported. The README asks for read-only database users and read-only keys throughout.
The app itself is open source, but it is not self-contained. It needs a Thesys API key: the Build agent's model calls go through Thesys' gateway (the default model is anthropic/claude-sonnet-5.5), and dashboards and chats are stored on OpenUI Cloud. You do not need an Anthropic key, though you can bring your own OpenAI or Anthropic key through the Thesys console. Thesys also runs a hosted copy at opendashboards.com.
Is OpenDashboards worth it?
OpenDashboards is best read as a reference app for Thesys' OpenUI Dashboards API rather than a standalone open-source BI tool. The code you can fork handles connections, credentials, the Build agent's tool loop and the live data route; generating the dashboard, storing it and storing the chat history happen on OpenUI Cloud. Thesys pitches it as an open alternative to Claude Dashboards (now Claude Data), and the honest comparison is that it swaps a closed Anthropic product for a closed Thesys service with an open shell around it.
The shell is well made. In our reading of the code, every host a user types is resolved and checked inside the socket's own DNS lookup, so only public addresses are reachable and cloud metadata endpoints, private ranges and DNS rebinding tricks are refused in every environment. Connection secrets are sealed with AWS KMS in production, which refuses to store them unencrypted; only local development keeps them in plain text. Queries carry hard limits: 15-second statement timeouts on Postgres and MySQL, 200 rows per exploratory query for the agent, 10,000 rows per dashboard query, and a 10 GiB billed-bytes cap per BigQuery query. Tool calls take the user and app from the session, never from the model's arguments.
Where your data goes depends on the step. When a dashboard opens, its queries run on your server and results go straight to the browser; OpenUI's docs say the cloud stores each dashboard's definition, not its data. While building, though, the agent runs exploratory queries and their results, up to 200 rows at a time, travel through Thesys' gateway to the model. Thesys promises zero data retention on paid models and warns that free models may use data for training.
The catch: self-hosting is narrower than it sounds. The address guard has no exception, so a database on a private network or on localhost cannot be connected without changing the code. Production assumes Supabase auth (you write your own AuthProvider otherwise) and an AWS KMS key reached through a Vercel OIDC role, so a plain Docker deployment means adapting both. Running OpenUI Cloud inside your own network is only on Thesys' custom-priced Scale plan. The project is two weeks old, at version 0.1.0 with no tagged release, 64 commits from 8 contributors, and the comparison table's "up to 10x faster" claim against Claude Dashboards comes without a method.
Our take: OpenDashboards suits a startup whose numbers already live in a cloud Postgres, PostHog or BigQuery and that is happy to pay Thesys per call; teams who need dashboards that never leave their network are still better served by Grafana or a classic BI tool with a hand-written query.
Where it earns its place
A founder at a ten-person SaaS company with a Neon Postgres database and Stripe
See MRR, churn and signups by source without writing SQL or hiring an analyst
Sign up at opendashboards.com, add a read-only Postgres user and Stripe's MCP server, then ask for a revenue dashboard. It refreshes on every open; follow-up messages change single charts. Thesys' free tier covers 3K API calls a month.
A growth marketer at an e-commerce brand
Combine Google Analytics 4, Search Console and PostHog into one weekly channel view
Connect each with a service account key or a read-only personal API key and ask one question across them. The agent checks the numbers with sample queries before building, so expect some of that data to reach the model through Thesys' gateway.
A platform engineer at a company already paying for Thesys C1
Give internal teams a chat-to-dashboard app on the company's own infrastructure
Fork the repo, write an AuthProvider for your identity provider, set up an AWS KMS key and deploy. Data sources must be publicly reachable, and keeping OpenUI Cloud itself inside the network means the custom Scale plan.
Before you commit
- Setup
- Hosted: sign up at opendashboards.com. Self-hosted: Node 20.9+, pnpm and Docker, a Thesys API key from console.thesys.dev, then clone, add THESYS_API_KEY, run pnpm db:up, pnpm db:reset and pnpm dev for a local app at localhost:3000 running as a dev user.
- Integrations
- PostgreSQL, Supabase, Neon, MySQL, MariaDB, ClickHouse (HTTPS), BigQuery, PostHog, Google Analytics 4 and Search Console built in; any remote MCP server with API-key auth. Uses the OpenUI server and React packages and AG-UI events.
- Data and privacy
- Credentials stay in the app's own Postgres, sealed with AWS KMS in production. Live dashboard results go from your server to the browser; dashboards and chats are stored on OpenUI Cloud, and the agent's exploratory results (up to 200 rows per query) go through Thesys' gateway to the model. Thesys offers zero data retention on paid models; free models may use data for training.
- Cost at scale
- OpenUI Cloud Build plan $49/month for 25K API calls ($0.002 per extra call) or Grow $499/month for 500K ($0.001 per extra call), plus model tokens at provider prices; private OpenUI Cloud is custom-priced. The app's own hosting and Postgres are small.
- Limits
- Build agent capped at 24 model rounds per turn; 15-second Postgres and MySQL statement timeouts; 200 rows per agent query and 10,000 per dashboard query; 10 GiB billed per BigQuery query. Only public hosts are reachable; MCP sources must be HTTPS with no OAuth.
What’s good
- Live dashboards that re-run their queries on open, with per-chart edits by chat
- Seven built-in connectors plus remote MCP servers for tools like Stripe and HubSpot
- Strong SSRF guard checked at DNS lookup, blocking private ranges and metadata endpoints
- Credentials sealed with AWS KMS in production; plaintext refused outside local development
- Row, timeout and BigQuery cost caps on every query
What’s not
- Requires a Thesys API key; dashboard generation and storage run on OpenUI Cloud
- Agent's exploratory query results pass through Thesys' gateway to the model
- Cannot reach databases on private networks or localhost without code changes
- Production wiring assumes Supabase auth, AWS KMS and Vercel OIDC
- Two weeks old, version 0.1.0, no tagged releases
Best for
Small product or growth teams with data in cloud Postgres, BigQuery, ClickHouse or PostHog who want chat-built live dashboards and accept a Thesys subscription.
OpenDashboards alternatives
OpenDashboards: common questions
- Is OpenDashboards worth it?
- A careful, well-guarded front end for asking your database questions and getting live charts back, but the part that builds and stores the dashboards is Thesys' paid cloud, so "open source" covers less than the name suggests. We rate it 3.5 out of 5.
- How much does OpenDashboards cost?
- MIT licence. Running it needs a Thesys key: OpenUI Cloud is free for 3K API calls a month, Build $49/month for 25K, Grow $499/month for 500K, with model tokens billed separately; self-hosting OpenUI Cloud is on the custom Scale plan. Hosted opendashboards.com pricing is not published.
- What are the downsides of OpenDashboards?
- Requires a Thesys API key; dashboard generation and storage run on OpenUI Cloud; agent's exploratory query results pass through Thesys' gateway to the model; cannot reach databases on private networks or localhost without code changes; production wiring assumes Supabase auth, AWS KMS and Vercel OIDC; two weeks old, version 0.1.0, no tagged releases.
- Who is OpenDashboards best for?
- Small product or growth teams with data in cloud Postgres, BigQuery, ClickHouse or PostHog who want chat-built live dashboards and accept a Thesys subscription.
- How much does OpenDashboards cost for a team or in production?
- OpenUI Cloud Build plan $49/month for 25K API calls ($0.002 per extra call) or Grow $499/month for 500K ($0.001 per extra call), plus model tokens at provider prices; private OpenUI Cloud is custom-priced. The app's own hosting and Postgres are small.
- Is OpenDashboards safe for company data?
- Credentials stay in the app's own Postgres, sealed with AWS KMS in production. Live dashboard results go from your server to the browser; dashboards and chats are stored on OpenUI Cloud, and the agent's exploratory results (up to 200 rows per query) go through Thesys' gateway to the model. Thesys offers zero data retention on paid models; free models may use data for training.
- What are the best OpenDashboards alternatives?
- The closest alternatives we have reviewed are Grafana, AG-UI and NocoDB.
Answers checked Nov 20, 2026.