Workspace limits and features

Most of what Phases can do is switched on for you already. The parts that are not are switched by a named control an organization admin can see, and the allowances your workspace is measured against are numbers you can read. This page is the reference for both: what exists, what it is set to, and what changes when it moves.

Where an admin sees this#

Settings → Feature flags, at /ws/<slug>/settings/flags. The page is headed Workspace features and opens with three counters — Enabled values, Custom overrides and Enforced globally — followed by a search box (Search by name, key, or description…) and a state filter offering All states, Enabled, Disabled, Custom overrides and Enforced globally.

The tab appears only for an admin of an organization workspace. A personal workspace has no Feature flags tab, and opening the URL directly redirects back to Settings — the values still apply to you, they are simply the platform defaults, and the Usage tab described below is where you see their effect. An organization member who is not an admin and navigates straight to the page gets it read-only, with Workspace admin access is required to change this value. under each control.

The allowances#

Two entries are numbers rather than switches, and they are the ones people ask about first.

Allowance	Key	Personal	Organization
Monthly Simulation Runs	`limits.max_runs_per_month`	100 runs	500 runs
Workspace Storage	`limits.storage_mb`	500 MB	5000 MB

A workspace admin cannot change either. The card carries no switch, only the line Numeric allowances are managed by a platform admin.

You see them in use on Settings → Usage, which draws three meters — Simulation runs, Messages and Storage. Simulation runs reads 12 / 100, Storage reads 48.3 MB / 500 MB, and the rail switches to a caution tone past 80% of the allowance. Messages has no limit attached and is a plain count.

Be clear-eyed about what that meter is. Nothing in the backend reads limits.max_runs_per_month or limits.storage_mb to refuse work: crossing 100 runs does not stop the 101st. They are displayed counters and a planning signal, not enforced caps.

Three kinds of control#

The rest of the page is boolean, split across three tabs.

Product access — Features and tools available in this workspace. What you and the agent can do.
Platform rollouts — Operational capabilities enabled for this workspace. How the work is executed, replayed and recovered. Every one of these is flagged High impact; they belong to whoever operates the deployment.
Usage limits — the two numbers above.

Reading one control#

Each control is a card with the same anatomy:

the friendly label (Material Knowledge Base) and its one-line description
the raw key in mono (feature.material_kb), which is what an error message or a support thread will quote
a Requires: Durable Simulation Jobs, Persistent Agent Sessions line when it depends on others, and a live warning — Not effective until Simulator Studio is enabled. — if you switch something on whose dependency is off
a risk badge of Use care or High impact (a standard control shows none)
the Effective value switch, reading Enabled for this workspace. or Disabled for this workspace.
a badge of Global default or Workspace override, an Enforced globally lock where a platform admin has blocked overrides, and a Use global default button that removes your override and returns the control to the platform value

Turning a High impact control on or off, or switching off something other enabled controls depend on, opens a confirmation first: This is a high-impact control. Saving creates a custom override for this workspace and takes effect immediately. — with the dependent controls named, and the raw key shown so you can check you are changing what you meant to.

What is on by default#

In the hosted product almost everything a user would ask for is already on. The table below lists the product-access controls a workspace starts with and the tools each one gates. The tool names are the ones in Tool reference.

Capability	Key	What it gates
Heat Treatment / Phases	`simulator.phases`	`run_ht_simulation`, the HT tab in the simulation panel
CCT Diagrams	`simulator.cct`	`run_transformation_diagram`
Resistance Spot Welding	`simulator.rsw`	`run_rsw_simulation`, `list_rsw_templates`
Virtual Gleeble	`simulator.vgleeble`	`run_vgleeble_simulation`
Material Knowledge Base	`feature.material_kb`	`list_materials`, `load_material`, `modify_material`, `compare_materials`, `save_material_to_workspace`
Flow Stress	`feature.flow_stress` (requires Material Knowledge Base)	`calculate_flow_stress`
Web Search	`feature.web_search`	`search_web` — current public sources, with citations
Python Analysis	`feature.python_analysis` (High impact)	`run_python_analysis`, derived analysis in a network-disabled sandbox
Engineering Analysis	`feature.engineering_analysis`	`compare_runs`, `run_analysis_recipe`
RSW Process Windows	`feature.rsw_process_windows` (requires Resistance Spot Welding)	`propose_rsw_process_window`, `approve_rsw_process_window`
Automatic Simulation Studies	`feature.simulation_studies` (requires Durable Simulation Jobs and Persistent Agent Sessions)	the Ultra Simulation toggle in the composer and the planned multi-trial study behind it
Simulator Studio	`feature.simulator_studio` (requires Persistent Agent Sessions)	`list_available_simulators`, `describe_simulator`, `run_simulator`, and the Simulator Studio entry in the left sidebar
AI Simulator Builder	`feature.ai_simulator_builder` (requires Simulator Studio)	the eight draft-authoring tools, from `list_simulator_operators` through `publish_simulator_draft` and `test_simulator_draft`
Equation and Model Runtime	`feature.simulator_model_runtime` (requires Simulator Studio)	approved algebraic, ODE and one-dimensional operators
Generated-code Sandbox	`feature.simulator_code_sandbox` (High impact, requires Simulator Studio)	`create_simulator_bundle` — digest-pinned generated solver bundles
Durable Custom Simulator Jobs	`feature.durable_custom_simulator_jobs` (High impact, requires Simulator Studio and Durable Simulation Jobs)	published simulator DAGs run through the durable coordinator

Two consequences are worth saying out loud, because the older documentation implied otherwise. Simulator Studio is live, so building and running a workspace simulator is something you can do today — see Custom simulators and Running a custom simulator. And the analysis and study tools are live: compare_runs, run_analysis_recipe and run_python_analysis are available in an ordinary workspace, as covered by Compare and analyse runs and Run a simulation study.

list_runs and inspect_run are gated by nothing at all. Reading back what a session already produced never depends on a flag.

Organization-only#

Two further controls are on for organization workspaces and absent from a personal one:

Workspace Invites (feature.invites) shows or hides the Invite Member form and the Pending invites table on Settings → Members. Turning it off takes the invite UI away; it does not revoke invites that are already out.
Team Collaboration (feature.team_collab) is catalogued and on, and nothing in the product currently reads it. Do not expect switching it off to restrict anything — sharing, membership and role enforcement are governed by Workspaces and roles, not by this flag.

What is off by default#

Four controls start off. One of them is a product capability, and the honest statement is that it is not something you can use today.

Control	Key	Status
Scientific Validation	`feature.scientific_validation`	Off. `inspect_validation_report` and `evaluate_validation_project` exist in the registry but are not enabled, so the agent answers `Scientific validation is disabled for this workspace.`
Missing Session Recovery	`feature.adk_missing_session_seed`	Off. A platform recovery path for agent sessions.
Route Control Plane	`feature.route_control_plane`	Off. Route-specific tool declarations and compact prompts.
Legacy In-process Execution	`feature.legacy_inprocess_execution`	Off. The compatibility path retained while durable jobs rolled out.

Nothing in the published validation record depends on that first control. What the solvers were checked against is in How we validate, Hardness validation and RSW validation.

Platform rollouts#

The remaining controls — Durable Simulation Jobs and the per-domain durable switches for HT/CCT, RSW and V-Gleeble, Durable Workflow Replay, Dynamic CCT/HT Workflows, Persistent Agent Sessions — decide how a run is submitted, replayed and recovered. They are on in the hosted product, which is why a simulation survives a closed tab and why a reconnect can replay narration from a stored cursor rather than a ten-minute memory buffer (Sessions, runs and artifacts explains that difference).

They are all High impact and several are each other's dependencies. Changing one alters the machinery under a working workspace, so leave them to whoever operates the deployment.

When something is switched off#

A disabled capability is reported, not hidden. A disabled simulator disappears from the simulator tab strip and its in-chat configuration card is replaced by a notice naming the capability. In conversation, the agent answers with the capability's own wording — for example Heat-treatment / phase simulation is disabled for this workspace. Ask your admin to enable it in settings.

Tool-level variants read the same way: Scientific validation is disabled for this workspace., Deterministic engineering analysis is disabled for this workspace., Staged RSW process-window planning is disabled for this workspace., Custom simulators are disabled for this workspace.

One caveat before you go hunting for a switch. A control makes a capability available to your workspace; a capability that needs a sandbox or an external worker also needs the deployment to provide one. If something stays unavailable after it is enabled, that is a question for whoever operates your installation rather than a flag you can fix.

A refusal that names a role rather than a capability is a different thing — This workspace role cannot run simulations. Ask a workspace admin for member access. means you are a viewer, and the fix is in Workspaces and roles. Common problems collects both kinds of message, and Reporting a problem covers what to do when neither explains what you are seeing.