Release notes

Newest first, dated by the day each change shipped. Beetl ships continuously, so there are no numbered releases.

November 2026

2026-11-09: Documents. A File Set now stores your files, and a file processor turns them into a dataset. CSV, NDJSON and Parquet files import automatically, as before. PDFs, Office files, images and other documents are read by AI into the fields you define, one row per document, and a review tab lets you correct what Beetl is unsure about. Excel workbooks can be mapped by cell ranges instead, with no AI. File processors appear on Home, and the assistant and MCP clients can read processors and their extracted records. File Sets no longer have processing options or an import status, and a changed schema no longer deletes and recreates the dataset. See Documents and spreadsheets and File Sets.

2026-11-03: Connect SDK 0.6 and write mode. Integration packages built before SDK 0.5 are rejected and must be rebuilt with the current SDK. You now choose a sync's write mode, Append or Replace, in Beetl for each sync, and Merge is no longer available for syncs. An Append sync continues from its last checkpoint, and a Replace sync starts over and writes the full output. See Build integrations.

2026-10-27: Standalone apps. An app can open full screen at its own address, with an assistant beside it that can change only that app's own settings and inputs. Right-click an app on Home and choose Copy standalone URL to share the address with colleagues. See Apps.

October 2026

2026-10-23 — Reports. Reports build PDF documents from your data with a Typst layout and SQL bindings, on demand or on a schedule, and appear as nodes on Home. Automations list their findings and can render them through a report template as a PDF. The assistant shows reports and data-flow views beside the chat, and MCP clients can read and manage reports and trace lineage. See Reports and Automations.

2026-10-21 — Apps. Apps are applications Beetl provides. They read your datasets, run on a schedule or when a dataset updates, and publish results back as datasets. Pallet Planner is the first. The assistant can operate app screens, with your confirmation for changes, and MCP clients can read apps and their runs. See Apps and Tool reference.

2026-10-20 — Schedules follow daylight saving time. Automation and sync schedules run correctly across clock changes, and timezone fields have a searchable picker. The default managed model is now Claude Haiku 5.5. See Automations.

2026-10-12 — Notebooks over Flight SQL. Python notebooks can query your datasets directly with a read-only Flight SQL key created under Settings → User.

2026-10-09 — Monthly usage. Settings → Tenant → Usage shows the month's sync runs, pipeline minutes and AI credits against your plan.

2026-10-05 — Long chats are summarized. When a conversation nears the model's context window, earlier turns are summarized, and a context meter in the composer shows how much is used.

2026-10-01 — The assistant gets tenant instructions and a stop button. A TenantAdmin can write up to 4,000 characters of workspace conventions under Settings → Tenant → AI, and the assistant reads them on every turn. Chat turns can be stopped while they run. See AI in a tenant.

September 2026

2026-09-29 — Managed AI. A new tenant gets working AI with no setup, paid from monthly plan credits. A tenant can bring its own key instead, and users choose the model and effort per chat session and per automation. See AI in a tenant.

2026-09-24 — External datasets. A TenantAdmin can register a PostgreSQL database, and users register its tables as datasets that are queried live, with nothing copied. See External datasets.

2026-09-22 — Automations over MCP. MCP clients can create, run and read automations, and connect at https://app.beetl.io/mcp. A read key now lists only the read tools. See MCP setup.

2026-09-15 — Registered datasets. Delta, Parquet, CSV or JSON files already in your object store can be registered and queried in place. See Registered datasets.

2026-09-10 — Integrations run in Beetl's sandbox. Every build, connection test, OAuth step and sync of a Connect package runs in an isolated sandbox that can reach public hosts over HTTPS. See The sandbox.

2026-09-08 — Ingestion rebuilt. Connection Types, Connections, Configured Syncs, File Sets and Output Links replace the old connection model. Webhook is the only built-in connection type; every other system arrives as a Connect package. The older built-in connection types (PostgreSQL, HTTP Feed, Server-Sent Events, Business Central and SAP Business One), the edge connector and file connections were removed. Files now go into File Sets, up to 5 GiB per file. Connections can be edited after creation, syncs have their own schedules, and sync runs have history and logs. Existing connections were not migrated; their datasets were kept. See Connection types, connections and syncs.

August 2026

2026-08-27 — Bring your own object storage. A TenantAdmin can register the tenant's own S3-compatible buckets under Settings → Tenant → Object storage. Beetl's hosted store remains the default. See Use your own bucket.

2026-08-25 — The assistant edits the page you are looking at. The assistant can read and change what is on screen in Query, Home, the pipeline editor and the automation editor. Edits flash where they landed and can be reverted for the turn. See Building with the assistant.

2026-08-25 — API key scopes collapsed from three to two. read_only and draft both became read; production became write. Existing keys keep working.

2026-08-20 — Scheduled automations run on time. A fault that could hold back scheduled runs was fixed. Each run now reads a shortened version of the previous run's report rather than its full transcript.

2026-08-19 — A simpler sidebar, and optional pipeline destinations. The sidebar is now Home, Chat, Automations, Dashboards and Query, and connections, datasets and pipelines are managed from the Data Landscape graph on Home. Separately, a pipeline can now have no destination at all, which makes a sources-and-stages exploration pipeline a first-class thing.

2026-08-18 — Tags on connections, datasets and pipelines. Tags are set per entity and filtered from the graph toolbar on Home. See Data Landscape.

2026-08-16 — Automations shipped. A recurring, observe-only investigation described in plain language, with a cron schedule, a test-before-save run, and a report that links every query it ran to the result behind it. Automations belong to the workspace rather than to the person who created them. See Automations.

2026-08-14 — The previous interface was retired. The app at /web is now the only interface.

2026-08-13 — New sign-in. You now sign in on a dedicated sign-in page, and signing out ends the session there as well. The app itself has no password fields.

2026-08-13 — MCP for your own clients. Each user creates their own MCP API keys, and each key has a scope that decides which tools a client can call. See MCP setup.

2026-08-11 — "Connector" was retired from the UI. "Integration" is the official term for a custom connector, and "Connection" remains the link to one source system. See Connection types, connections and syncs.

2026-08-05 — The assistant can read a wiki. The assistant can read and search the platform reference documentation and a generated page for each deployed pipeline. It cannot write to them.

July 2026

2026-07-30 — Per-tenant object stores. Each tenant has its own object stores. Datasets stay in the store they were written to, and changing the default store does not move existing data.

2026-07-24 — The assistant became persistent. A collapsible right-hand panel on every page keeps one ongoing assistant conversation per user, and the page you are on can share what it shows with the assistant.

2026-07-20 — A new app. The product moved to a new interface, with the Data Landscape graph as the home page and embedded Superset dashboards as a new section.

Earlier

2026-06-03 — Chat asks before it deploys. Before the assistant deploys a pipeline, chat shows an Allow or Deny card, and nothing deploys until you choose Allow.

2026-06 — Pipeline variables and run context. A Variables panel for vars.*, a run-values dialog for run.*, fn.today() / fn.now() / fn.uuid() helpers, and per-source trigger mapping. See Template variables.

2026-05-23 — One file, one connection, one dataset. File upload was rebuilt around explicit user-supplied dataset names, a schema-checked Replace operation, and a clear refusal of unsupported formats.

2026-05 — Saved query results, charting, and lineage. Every query result is saved for 7 days with its own URL, can be charted, and shows where each cell's value came from, with drill-down to the records behind it.

2026-05 — Server-Sent Events connections. Beetl could subscribe to an event stream and write its events to a dataset. This connection type was removed on 2026-09-08.