# Release notes (/release-notes)
What changed in Toucan AI, newest first. Each month lists the new features, improvements and notable fixes that shipped to production during that month.
The current month is updated as changes ship and finalized at the start of the following month.
## September 2026 [#september-2026]
This month is still in progress: this section covers changes shipped up to 23 September 2026.
### Highlights [#highlights]
* **Semantic domains** — measures and dimensions can now be owned by a business domain, and a database can be set to a *semantic-only* referencing mode so the AI only uses curated semantic objects.
* **Schema refresh** — refresh a database's schema from the database page. Toucan AI merges the new structure with your existing configuration (visibility, descriptions, RLS) instead of starting over.
* **Pilot role** — a new organization role. See [Organization roles](/govern/roles-and-permissions/organization-roles).
* **Claude Opus 5 and Claude Sonnet 5** are available in the AI provider and model selection. See [Choose your AI provider and model](/govern/ai-control-and-optimization/select-your-ai-provider).
### New features [#new-features]
* **Materialized views** can be included when connecting Oracle, PostgreSQL and Snowflake databases.
* **Database connection status** is shown on each database, so broken connections are visible at a glance.
* **Data step suggestions** during AI analysis propose cleaning and preparation steps for your tables.
* **Remote RLS for PostgreSQL** — Toucan AI can pass session variables to your PostgreSQL database so your own row-level security policies apply.
* **SSO password exemption** — on a verified SSO domain, specific users can still be allowed to sign in with a password.
* **Custom home page** — an organization can replace the default home page with a custom front.
* **Custom fronts** (early access): embed-script support, a maker debug panel showing what each SDK call returned, a confirmation before switching an organization's access, and a generated SDK reference for coding agents.
* **Object access dialog** to review and manage who can access a given object.
### Improvements [#improvements]
* Table AI-readiness is computed server-side, making large database pages faster.
* Advanced connection settings are grouped in their own section of the database editor.
* Navigation is reachable on small screens.
### Fixes [#fixes]
* The API now answers `401 Unauthorized` (instead of `403`) when no valid credential is provided.
* Oracle sample-data queries are bounded to avoid timeouts.
* A failed organization creation is rolled back and the error is shown.
* The subscription plan is now derived correctly from the subscribed products.
## August 2026 [#august-2026]
### Highlights [#highlights-1]
* **Organization SSO with SAML and OIDC** — connect your identity provider so members sign in with their company account. See [Single Sign-On (SSO)](/govern/single-sign-on).
* **New AI analysis review** — AI analysis now runs as tracked *runs*. Each run produces proposals with a confidence level and a rationale that you review and confirm, from a run list and a dedicated review page. See [Review AI analysis proposals](/build/analyze-your-database-with-ai/how-to/review-ai-analysis-proposals).
* **Self-service dashboards for Explorers** — platform Explorers can build their own dashboards.
* **Custom fronts** (early access) — host your own front-end inside Toucan AI, with an SDK (including `getUserInfo()`), a key-value store (shared and per-viewer), SSO authentication, attribute overrides for previewing as another user, and an AI assistant panel the front can open.
### New features [#new-features-1]
* **Requests tab** — raise requests to the Toucan team directly from the product and follow their status.
* **AI-suggested relationships** between tables, with names and descriptions on foreign-key relationships.
* **Domain and Data screens** in the semantic layer (early access).
* **Maker MCP** — makers can connect with OAuth instead of an organization API key, and manage custom fronts through the MCP.
### Improvements [#improvements-1]
* The AI assistant receives table relationships when searching for tables, producing better joins.
* Clearer message in the AI chat when the organization has no credits left.
* Better agent feedback for unusual column names and missing items.
* Side tabs on the Data and Domains pages; general design refresh.
### Fixes [#fixes-1]
* Duplicated dashboards stay unpublished until you publish them.
* Columns whose names match JavaScript built-ins (e.g. `constructor`) are read correctly.
* Database connection pool timeouts under load.
## July 2026 [#july-2026]
### Highlights [#highlights-2]
* **Explorer role** — a new role for end users who consume and explore content, with their own user attributes. See [Organization roles](/govern/roles-and-permissions/organization-roles).
* **Text blocks** in dashboards, to add titles, explanations and context next to charts.
* **Relationships** — define relationships between tables (with a name and description), created automatically during analysis and editable afterwards.
* **Table-level data preparation** — cast columns to another type and apply pipelines at the table level, so every chart built on the table benefits. Table preview and preparation are merged into a single *Data* tab.
* **Organization Memory** — the AI keeps durable, organization-level knowledge it learns from conversations.
### New features [#new-features-2]
* Measures and dimensions can be defined on a database's tables (semantic layer).
* Column-Level Security can be edited from the database page.
* AI analysis suggests validated column type casts.
* New `datetimefromparts` data step.
* **Free-trial onboarding** on the home page, with a product-help agent that answers questions and can connect sample data for you.
* **Fill Organization Context with AI** — look up your company to pre-fill the organization context.
* **Multi-session account switcher** to move between accounts without signing out.
* New dashboard list with embed cards in the Library.
* The AI assistant can edit every dashboard filter type, and suggests contextual prompts when editing a chart or filter.
* Oracle: optional schema (owner) for introspection and `SSL_SERVER_CERT_DN` option.
* hCaptcha protection on sign-up.
### Improvements [#improvements-2]
* The AI assistant generates semantic objects (measures and dimensions) instead of raw queries.
* Charts check readability and report facts deterministically before being submitted by the AI.
* Tables without complete embeddings are flagged as not AI-ready; database cards can be filtered by analysis status.
* Column type icons in data previews; the compact date range picker shows years properly.
* Chart narratives are capped at 200 characters.
### Fixes [#fixes-2]
* Column-Level Security is applied to AI assistant tools.
* Conversations are strictly isolated between organizations.
* NULL dates are no longer displayed as January 1, 1970; zero values are kept when excluding missing values.
* Legends no longer overlap labels; chart resize handles are more visible.
* Charts show an empty state instead of an error while being configured.
## June 2026 [#june-2026]
### Highlights [#highlights-3]
* **Semantic chart editor** — build charts from metrics and dimensions defined in the semantic layer, with metric composition and filters on dimensionless charts. See [Semantic Layer](/build/semantic-layer).
* **New chart types** — area line chart, stacked area chart and bubble chart. Heatmaps get color options. See [Chart types](/build/charts/chart-types).
* **Export chart data** as CSV or XLS.
* **Per-organization AI provider and model** — choose the AI provider and model used by your organization. See [Choose your AI provider and model](/govern/ai-control-and-optimization/select-your-ai-provider).
* **AI context v2** — the AI context is organized in four configurable sections. See [Adding Context to the Agent](/govern/ai-control-and-optimization/how-to).
### New features [#new-features-3]
* **Credit limits per user** — organization admins can cap how many credits each user can consume. See [Credits & Usage](/getting-started/credits-and-usage).
* **Embed dashboard builder** modal to generate embed code for a dashboard.
* Edit dashboard charts and filters directly from the AI conversation panel.
* "Improve with AI" for metric definitions.
* Dashboard search bar in the Library.
* Sub-day granularities (hours, minutes) for date granularity and time shift steps.
* Redesigned sign-up and sign-in pages; SSO sign-up.
* Oracle TLS wallet upload.
* New supported models: GLM 5.2, Mistral Medium 3.5 (via Scaleway).
### Improvements [#improvements-3]
* Simpler, single-agent AI architecture: faster and more reliable answers, with more turns allowed per request.
* Guardrails ensure SQL queries are never exposed in AI answers.
* Clearer chart error messages and clearer errors for SQL syntax used in semantic expressions.
* The *Clear chat* action moved to the prompt toolbar; only the latest feedback toolbar is shown.
* A warning is shown before refreshing a database schema.
* Faster schema loading for databases with many schemas.
* The "beta" mention was removed across the product and emails.
### Fixes [#fixes-3]
* Leaderboard values no longer overlap labels.
* Joins rename duplicate columns automatically.
## May 2026 [#may-2026]
### Highlights [#highlights-4]
* **Oracle connector** — connect Oracle databases. See [Supported databases](/build/data-connections/supported-databases).
* **Plans and billing** — subscriptions, with features available depending on your organization's plan. Credit tooltips link to the billing documentation. See [Credits & Usage](/getting-started/credits-and-usage).
* **Encrypted connection credentials** — database connection information is now encrypted at rest. See [Secrets & encryption](/govern/security-model/secrets-and-encryption).
### New features [#new-features-4]
* Drag and drop to reorder dashboard filters.
* Semantic charts: display and support for multiple metrics (foundation for the semantic chart editor).
### Improvements [#improvements-4]
* Long conversations are summarized automatically, so the AI keeps context without hitting limits.
* More flexible AI planning for multi-step questions.
* The AI uses the current date for relative date questions ("last month", "this year").
### Fixes [#fixes-4]
* Editing a chart creates a new dashboard version.
* Dashboard chat history is no longer lost.
* The chain of thought is visible again in the AI chat.
* MySQL: looser table-type compatibility check.
* A pending subscription no longer blocks checkout.
## April 2026 [#april-2026]
### Highlights [#highlights-5]
* **Snowflake and ClickHouse connectors.** See [Connect a Snowflake database](/build/data-connections/how-to/connect-a-snowflake-database).
* **Chain of thought in the AI chat** — see the steps and query pipelines the AI runs to answer your question.
* **Dashboard versioning** — dashboards keep a version history, like charts.
* **Credits** — credits used are computed and shown on AI usage. See [Credits & Usage](/getting-started/credits-and-usage).
* **Embeddable self-service dashboard** and **Pin to dashboard** components. See [Embed a self-service dashboard editor](/getting-started/quick-start/embed-a-dashboard-1).
### New features [#new-features-5]
* Follow-up question suggestions after a chart answer.
* Dashboard filter variables are used in query execution.
* Tabs for the different AI chat embed snippets.
* Wildcards supported in authorized CORS origins.
* KPI values can use their own default dimension, independent of other charts.
* Larger data preview limit.
* Oracle listed as coming soon.
### Improvements [#improvements-5]
* Table analysis uses vector embeddings to find the relevant tables for a question, with a fallback to classic metadata search.
* Clearer error messages for embed tokens, prompt sending and failed chart generation.
* Improved dashboard editing experience.
* Column types are unified across the product.
### Fixes [#fixes-5]
* Column-Level Security is applied on database endpoints.
* AI answers no longer leak between dashboards while streaming.
* Dates before 1970 are supported in chart date ranges.
* Table visibility is kept after analysis and RLS changes.
## March 2026 [#march-2026]
### Highlights [#highlights-6]
* **MySQL connector.** See [Supported databases](/build/data-connections/supported-databases).
* **Column-Level Security (CLS).** See [Column-Level Security](/build/security-and-governance/column-level-security-cls).
* **SSO sign-in options** on the authentication pages.
* **AI usage page** — admins can see how many tokens their organization consumed, also summarized on the home page.
* **New editing experience** — charts are edited in the context of their dashboard and are no longer shared across dashboards.
### New features [#new-features-6]
* **User model** with default attribute values, replacing user variables in settings.
* **Table visibility** — choose which tables are visible, with search and scroll on the tables page. See [Table Visibility Governance](/govern/table-visibility-governance).
* **Step explanations** in the AI chat.
* Dynamic starter suggestions in the AI chat; your draft prompt is kept if you leave the page.
* Welcome form after sign-up and additional getting-started steps.
* Database suggestions when adding a database.
* API keys are now scoped to the organization. See [Generate an API key](/embed/authentication/how-to/generate-an-api-key).
* Support section with links to the status page and documentation.
* The date range filter is available in French.
* AI thinking messages are translated.
### Improvements [#improvements-6]
* Better chart axis labels; dates displayed in the user's locale.
* Revamped table status indicators.
* Unsaved-changes warning on the database tables page.
* Redesigned emails, new logo and favicon.
* Value chart polish.
### Fixes [#fixes-6]
* Percentages are expected between 0 and 1.
* Tooltips stay inside the chart container.
* Clearer error when a database has been removed.
* Reduced AI hallucinations on insights.
## February 2026 [#february-2026]
### Highlights [#highlights-7]
* **Dashboard publishing** — work on a draft and publish when ready; embedded and shared dashboards show the published version.
* **RLS operators** — richer Row-Level Security rules, also applied to filter values. See [Row-Level Security (RLS) operators](/build/security-and-governance/row-level-security-rls-operators).
* **Custom AI instructions** at the organization level. See [Set AI instructions](/govern/ai-control-and-optimization/how-to/set-ai-instructions).
* **Home page with a getting-started checklist**, a sample dataset for onboarding and a feedback form.
* **New AI architecture** — dedicated agents for charts, queries and dashboard filters.
### New features [#new-features-7]
* Dynamic AI context per embed token via the `aiContextClues` field.
* Schema selection before AI analysis; the analysis opens automatically after adding a database.
* Database structure updates are supported.
* Query preview and schema selector in the source form.
* See in which dashboards a chart is used; clearer confirmation when publishing a chart.
* AI chat suggestions and a friendlier "thinking" display.
* New charts are placed at the bottom of the dashboard with auto-scroll.
### Improvements [#improvements-7]
* New design system across the product; standardized page headers.
* Improved autosave and reduced flickering on dashboards.
* Dates and timestamps are formatted in data previews; alternating rows in tables.
* Updated AI chat and embed instructions UI.
### Fixes [#fixes-7]
* Views are scoped to user permissions.
* Charts handle a removed database gracefully.
* Embedded dashboards can scroll.
* Dashboards refresh when a chart is created by the AI assistant.
## January 2026 [#january-2026]
### Highlights [#highlights-8]
* **Chart autosave and live preview**, with a **version history** panel to see and restore previous versions.
* **Chart publishing** — see when a chart was last published from the Library and the history panel.
* **Theming** — customize colors and appearance, including in embeds. See [White-Labelling & Branding](/embed/white-labelling-and-branding).
* **Variables** with default values. See [Variable management](/build/variable-management).
* **New chart types** in the manual editor: stacked bar chart and heatmap.
### New features [#new-features-8]
* Duplicate charts and dashboards; add an existing chart to a dashboard.
* Embed a single chart by id (view mode).
* Charts generated by the AI are applied automatically, with a suggestion to open the chart editor.
* AI chat: welcome messages, dates in history, thumbs up/down and copy on messages, and an "AI can make mistakes" disclaimer.
* A dialog explains how to embed the AI chat.
* Language selector; chart types are translated.
* Dates aggregated by granularity.
* IP addresses to whitelist are suggested in the database form.
* "Visualizations" is renamed **Library**; updated sidebar navigation and account settings.
* Public documentation moved to GitBook.
### Improvements [#improvements-8]
* Embedded components no longer inject CSS into host websites.
* New empty states in the Library, chart editor and dashboards.
* Chart conversation history is isolated from the dashboard's.
### Fixes [#fixes-8]
* Heatmap configuration works correctly.
* Dashboards autosave when adding or removing charts.
* Invitations work again.
## December 2025 [#december-2025]
### Highlights [#highlights-9]
* **New chart and dashboard editing experience**, with a resizable editor panel and reworked navigation.
* **Conversation history** — AI chats are saved and can be resumed.
* **New chart types** in the manual editor: horizontal bar chart with ranking, and circular chart.
* **Self-service for Explorers in embeds** — embedded users can create a dashboard from scratch and reload the ones they created.
### New features [#new-features-9]
* Charts can include a narrative.
* Join and append steps in chart queries; new date granularity step.
* Common aggregation functions, limited to those supported by the column type.
* Y-axis range, date format selection and an option to exclude null values.
* New query editor.
* Embed permissions can target all objects of one type. See [Permission level overview](/embed/permissions-and-row-level-security/permission-level-overview).
* AI metadata enrichment for BigQuery.
### Improvements [#improvements-9]
* Reworked databases list, database form and database tables pages, including RLS.
* Chart options grouped in accordions (bar, line, table and value charts).
* Embeds use light mode by default.
### Fixes [#fixes-9]
* Value charts can be edited manually.
* New charts can be saved from the chart editor.
## November 2025 [#november-2025]
### Highlights [#highlights-10]
* **The AI assistant can generate and edit charts.**
* **Google BigQuery connector.**
* **Row-Level Security** — configure and apply RLS, with a preview. See [Apply RLS to your database](/embed/permissions-and-row-level-security/how-to/apply-rls-to-your-database).
* **Embeddable AI conversation.** See [Embed an AI Chat](/embed/embedding-overview/how-to/embed-an-ai-chat).
### New features [#new-features-10]
* Build charts without AI using a form: pick columns from the query output, pre-built aggregations, sorting and number formatting.
* Edit a chart's source from the chart editor; the chart configuration is kept when the new source allows it.
* JSON editor for query steps.
* Dashboard filters work on all columns used in a chart.
* Chart library and a new Visualizations page.
* Collapsible AI assistant chat.
### Fixes [#fixes-10]
* Embedded dashboards respect dark and light mode.
* Filter popovers render correctly inside embeds.
## October 2025 [#october-2025]
### Highlights [#highlights-11]
* **AI assistant** — first version of the conversational assistant.
* **Dashboard filters** — filter values are fetched from chart sources and applied to chart queries.
* **Value (KPI) chart.**
* **Embed token generation UI.** See [Generate a token](/embed/authentication/how-to/authentication-and-tokens).
### New features [#new-features-11]
* Dark mode for charts; colors follow the theme palette.
* Number formatting (unit, precision) and custom labels for column names.
* New table chart and improved data zoom.
* Database connection status.
* Query validation.
### Improvements [#improvements-10]
* Faster data transfer between services.
* Empty states, loading skeletons and first Toucan styling.
## September 2025 [#september-2025]
### Highlights [#highlights-12]
* **Dashboards** — create dashboards, add, edit and delete charts in a grid layout.
* **Embedding** — embed a dashboard in your product with a script. See [Embed a dashboard](/embed/embedding-overview/how-to/embed-a-dashboard).
* **API keys and embed tokens**, with permissions and token attributes. See [Token-based access](/embed/authentication/token-based-access).
### New features [#new-features-12]
* Authorized CORS origins per organization.
* Chart generation with AI.
* First filtering components.
### Fixes [#fixes-11]
* Embed CSS no longer conflicts with the host page.
## August 2025 [#august-2025]
First release of Toucan AI.
### Highlights [#highlights-13]
* **Accounts and organizations** — sign up, create an organization, invite members, reset your password.
* **PostgreSQL connections** — add, test, edit and delete database connections with a guided wizard.
* **AI analysis** — Toucan AI analyzes your database structure and enriches it with metadata, with live progress.
### New features [#new-features-13]
* Table previews and database structure browser.
* Member administration and user settings page.
* English and French interface.
* First AI chart generation.