Workspaces

A Workspace is a Claude-style conversation grouping in Griffin AI Search. It bundles a system prompt, a default model, attached files, and a member list, so every conversation you start inside it shares the same context and configuration instead of you re-explaining it each time. Use a Workspace when a team or a project's worth of conversations should all draw on the same background — for example, a workspace pre-loaded with your compliance framework documents and a system prompt tuned for answering compliance questions.

What belongs to a Workspace

Only Name, Description, and System prompt are set in the initial Create a workspace panel (see below). Default model, Colour, and Skills are added afterward, from a created workspace's Edit tab — there's no way to set them at creation time.

A conversation belongs to at most one workspace at a time. When it does, the workspace's system prompt and default model apply to that conversation automatically — unless a Skill applied to the chat, or a setting on the conversation itself, overrides them.

Example: a workspace named "Q1 Compliance Review" with the system prompt You are answering questions about our SOC 2 Type II controls. Cite the specific control ID in every answer., default model set to griffin-2o, three attached policy documents, and a "3 members" badge — every conversation started under it inherits all four automatically.

Why use a Workspace

Without a workspace, every new conversation starts from a blank system prompt and whatever model you happen to pick. A Workspace lets you set that context once and have it apply everywhere it's used:

• Consistent framing — a system prompt tuned for a specific kind of question (compliance, incident response, a particular audit) applies to every conversation in the workspace, not just the one where you typed it.
• Consistent model — set a default model once instead of choosing it per conversation.
• Shared background material — files attached to the workspace are available as context without re-uploading them into each new chat.
• Shared access — anyone added as a member can use the workspace and its conversations, rather than context being trapped in one person's chat history.

Composing Skills into a Workspace

A Workspace can compose multiple Skills — reusable prompts — that apply to every conversation started in it. This lets you layer a workspace's baseline context (system prompt, default model, attached files) with one or more Skills for more specific behaviors, without having to re-select those Skills in each new conversation. Where a Skill's own instructions conflict with the workspace's system prompt or default model, the Skill takes precedence for the conversation it's applied to.

Skills are attached from a workspace's Edit tab, not the create dialog — you create the workspace first, then open it and check off which of your Skills should compose into it. For example, a "Q1 Compliance Review" workspace could have a soc2-controls Skill and a cite-sources Skill both checked, so every conversation started in that workspace runs with the workspace's own system prompt plus both Skills' prompts layered on top.

Creating a workspace

1. Click New workspace from the Workspaces tab, or select Create new workspace from the workspace picker in the conversation composer.
2. In the Create a workspace panel, fill in:
   • Name (required, e.g. Q1 Security Review)
   • Description (optional, e.g. Vendor risk questions for the Q1 audit)
   • System prompt (optional) — instructions applied to every chat in this workspace, e.g. Answer using our current vendor risk tiering. Flag anything above medium risk.
3. Click Create workspace.

The dialog description sums up the idea directly: "Group conversations under a shared system prompt and default model — like a project."

Managing workspaces

Workspaces are managed from the Workspaces tab. Since workspaces span products rather than belonging to one, the same workspace list is available regardless of which product you're working in. The table shows each workspace's name, description, default model, member count, and last-updated time, plus an enabled toggle and row actions to open a workspace's details or delete it.

Opening a workspace's details shows two tabs:

• Overview — a read-only summary: default model, member count, attached-file count, colour, created/updated timestamps, the full system prompt, and the list of composed Skills.
• Edit — where you can change the name, description, system prompt, default model, colour (with a colour-swatch picker), and which of your Skills are composed into the workspace. Click Save changes to apply.

If no workspaces exist yet, the tab shows: "No workspaces yet — create your first one to group related chats."

Attaching a workspace to a conversation

From the conversation composer, type > to open the workspace picker (this mirrors the / picker used for Skills). It lists your workspaces, showing each one's name and description, and filters the list by name as you keep typing — e.g. typing >Q1 narrows the list down to workspaces whose name matches "Q1", such as "Q1 Security Review." Select one to attach it to the current conversation, or choose Create new workspace to open the create panel without leaving the composer.

Member access

Each workspace has an owner and a list of members. Members have shared access to the workspace — its system prompt, default model, attached files, and composed Skills are available to them the same way they are to the owner. The Workspaces table and a workspace's Overview tab surface this as a member count badge (e.g. "3 members") rather than listing individuals inline.

Attached files

A workspace's data model has a slot for attached files, which would make them available as background context to every conversation in the workspace — useful for grounding a workspace in reference material like a compliance framework, a policy document, or a runbook, without attaching the same files to each conversation individually. A workspace's Overview tab shows how many files are attached. Neither the Create panel nor a workspace's Edit tab currently exposes a control for attaching files, so populating this today requires the API.

Common errors

"Workspace API not available yet" when trying to create a workspace — the backend workspace endpoints aren't available in your environment yet. Contact Safeguard support.

"Missing session context" when trying to create a workspace — your sign-in session is incomplete; reload the page and try again.

"Name is required" — shown both in the Create panel and when saving a workspace's Edit tab with an empty name.

A name over 80 characters, a description over 280 characters, or a system prompt over 8,000 characters is rejected — the Create a workspace panel validates these before submitting; shorten the field and resubmit. (These limits aren't enforced client-side when editing an existing workspace from its Edit tab.)

"Failed to create workspace" (Create panel) / "Failed to save workspace" (a workspace's Edit tab) / "Failed to update workspace" (the enabled/disabled toggle) / "Failed to delete workspace" — the request reached the backend but it returned an error; the toast's description shows the underlying message. Retry, and if it persists, contact Safeguard support.

A workspace's system prompt doesn't seem to apply — check whether the conversation has its own override, or whether a Skill applied to the chat is taking precedence; both can override the workspace's system prompt and default model for that conversation.

Related

• Griffin AI — the AI search and chat experience Workspaces organize conversations within.
• Skills — the reusable prompts a Workspace can compose into every conversation it contains, and the model for the / mention picker that Workspaces' > picker mirrors.
• Search — the Griffin AI Search interface whose composer hosts the > workspace picker.