DOCSEARCHbyAlgolia
Submit Docs

Speckle 2026.9 Documentation

website·Indexed 35 days ago

https://docs.speckle.systems/quickstart/welcome

Documentation for Speckle 2026.9, featuring a new language-neutral object model with a columnar property store and relational layer. Includes guides on the updated 3D viewer, Speckle Intelligence for AI-driven model analysis, and updated connectors for Revit, Grasshopper, Navisworks, Power BI, and Trimble Connect. Developer resources cover building integrations with the bundle builder, SGEO geometry encoding, and updates to GraphQL and REST APIs.

Tokens
39.5K
Snippets
39
Records
288
Agent score
64%

What's inside Speckle

  1. Overview of Speckle Intelligence AI Rules

    AI Rules are standing instructions that Speckle Intelligence follows in every conversation within a workspace. Unlike skills, which are opt-in via slash commands, rules are mandatory governance constraints (e.g., currency, language, safety, or disclosure) that apply to every turn of the Chat and the viewer sparkle. Only workspace admins can manage rules, though all members can view them in Settings or via the 'Workspace rules' control in Chat.

  2. Overview of SGEO Geometry Encoding in 2026.9

    In Speckle 2026.9, geometry is stored as a table of binary blobs (SGEO format) keyed by a geometry key and attached to objects via relations (DISPLAY, SOLID, or CENTERLINE). This replaces the previous method of storing geometry as displayValue JSON objects on DataObjects. Producers write SGEO blobs into bundles, while the server generates a .dat file for the WebGPU viewer.

  3. Overview of Speckle Automate in version 2026.9

    In Speckle 2026.9, the Automate interface has been reorganized. The previous 'Automations-tab' wizard has been removed. Functions are now managed at the workspace level, while specific automations (which link a function to a trigger model) are configured within project settings. Note that Speckle Automate is exclusively available on Speckle Enterprise Server.

  4. Overview of Speckle Intelligence

    Speckle Intelligence is an AI assistant integrated into the Speckle web app that allows users to ask questions about their published models in plain language. It provides answers, charts, and can manipulate the 3D viewer. It operates by running read-only queries against Speckle data rather than uploading raw host files (like Revit or Rhino) to an AI, ensuring efficiency and cost-control of AI tokens.

    Key capabilities include querying:

    • Object properties on published models.
    • Relationships between parts (containment, connectivity, levels, systems, materials).
    • Source application and data types of published versions.
    • Project update history (who published what and changes between versions).
    • Workspace library files and registered tables (CSV, XLSX, Snowflake, Databricks, Microsoft Fabric).

  5. Overview of the Speckle 2026.9 Relational Layer

    In Speckle 2026.9, the previous Proxy system (used for overlapping membership like materials, levels, and groups) is replaced by an explicit relational layer. This layer consists of typed, directed edges between objects and structural nodes. Proxies such as RenderMaterial, Level, Group, Definition, InstanceProxy, and InstanceDefinitionProxy no longer exist in 2026.9 bundle data; their functionality is now carried by specific relation types.

  6. Overview of Trimble Connect synchronization

    The Trimble Connect integration allows for automatic model synchronization. When a file is selected for sync, every new version uploaded to Trimble Connect automatically becomes a new version of a Speckle model.

    Key Mechanics:

    • Personal Browsing: Users sign in with their own Trimble ID to browse and select files; these credentials are not stored by Speckle.
    • Background Sync: Because Trimble Connect lacks machine-to-machine access, a dedicated Trimble service account must be connected to the Speckle workspace to handle background downloads.
    • Polling: Speckle checks for new file versions every 5 minutes.

  7. Overview of Intelligence skills in Speckle

    An Intelligence skill is a reusable playbook for Speckle Intelligence Chat. It allows organizations to standardize instructions—such as model coloring, pricing methods, carbon factors, or brand tone of voice—so they can be invoked across a workspace without being redefined in every conversation. Skills reduce AI token consumption by providing a consistent method for the assistant to follow.

    Skills are distinct from 'AI Rules', which are always-on constraints. Skills must be explicitly invoked using a slash command.

  8. Understand version metadata changes in Speckle 2026.9

    In Speckle 2026.9, bundle data no longer contains a root object. Scripts that previously relied on version.referencedObject to find a root collection or read model-wide facts (like units) will break. Metadata is now distributed across the server version record, the model ingestion, a provenance stamp in the bundle, and model-scoped properties.

  9. Understand the Speckle 2026.9 Object Model changes

    Speckle 2026.9 introduces a fundamental shift in how data is structured, moving away from the tree-and-Proxy model to a columnar property store and a relational layer. Key architectural changes include:

    • Membership as Relations: Membership is no longer a list of elements on a parent object. Instead, it is a relation (e.g., IN_COLLECTION or ON_LEVEL) set on the object itself.
    • Upfront Metadata: Source application and message metadata are provided during ingestion rather than being attached to a root object at the end.
    • Asynchronous Versioning: Publishing returns a reserved ID; the version only exists after ingestion completes. Developers must wait for the version to be finalized before querying, linking, or receiving it.

  10. Benchmark a project against a portfolio using Speckle Intelligence

    Speckle Intelligence Chat typically answers questions about one project at a time. To compare a current project against past work without manual exports, you can pin a benchmark library (containing figures like steel tonnage, embodied carbon, or cost per m²) as Knowledge. This allows Chat to analyze the current project's data relative to the pinned historical data within a single conversation.

    This workflow is ideal for flagging outliers and performing high-level comparisons. For real-time, side-by-side analysis of multiple live projects that update automatically, use the Portfolio Analysis feature in Dashboards instead.

  11. Understand the Speckle 2026.9 Integration Model

    In Speckle 2026.9, the data model for producers and consumers has shifted from a JSON-based object graph with a root object to a 'bundle' model. A version is now defined as a bundle (a set of parquet files), and the process of turning an upload into a version is called 'model ingestion'.

    Key architectural changes include:

    • Data Format: Shift from JSON objects to parquet files for objects, properties, geometry, and a graph of nodes/relations.
    • Object Identity: Use of applicationId per object. Objects, geometry, and nodes use dense integer keys local to a single bundle.
    • Hierarchy: Replaced nested elements and Proxy arrays with grouping containers, structural nodes, and typed relations.
    • Geometry: Replaced displayValue JSON objects with SGEO binary blobs in sharded geometry tables.
    • Version Creation: Instead of uploading objects and calling create-version, developers now open an ingestion, upload the bundle, and complete it. The version exists only when the ingestion reaches success status.

  12. Query project and model data with Speckle Intelligence

    Speckle Intelligence allows users to ask natural language questions about their BIM models. To query model data, versions must be in the new data format; versions in the old format cannot be queried.

    Common query examples:

    • Project-level: "What models are in this project, and who published recently?"
    • Version comparison: "Compare the two most recent versions and summarise what was added, removed, or changed"
    • Spatial/Relational: "What sits on this level, and what contains those objects?" or "What connects to this system?"
    • Quantities: "Give me a report of quantities by category and level" (Use the Chat interface for this).
    • Selection-based: "What is the history of the object I have selected?"