Humans and agents working on the same records. Shared Bases, threaded comments, approval flows, audit trails. Built so reviewers and auditors have everything they need.

Shared Bases

A Base can be private to one user, shared with a team, or shared cross-organization (e.g. your firm + your client’s counsel). Each member has a role that determines what they can see and do. Roles are defined per Base — a user can be admin in one Base and read-only in another.

Typical roles (customizable):

Role	Typical permissions
Admin	Everything — sources, agents, permissions, billing
Editor	Read + write on content, configure agents, run workflows
Reviewer	Read + comment + approve. Cannot modify content or agents.
Viewer	Read-only access to specified collections
External	Time-limited read access to a specific document or output

Comments & threads

Every row in a workflow output, every agent response, every document passage can host a threaded discussion. Comments can include:

• Plain text — the usual
• @mentions — notify a colleague, including via email or Slack if configured
• Agent invocations — “@MSA-Reviewer what’s your reasoning for this flag?” pulls the agent into the thread
• Attachments — additional context files

Comments are first-class audit artefacts. Every comment is timestamped, attributed, and immutable once posted (edits show edit history).

Approvals

Workflows can require human approval before specific steps execute. The approval can be:

• Inline — user clicks “Approve” in the UI
• Conditional — only required when a condition holds (amount > €100K, risk = HIGH, etc.)
• Multi-signature — requires N-of-M approvers
• External — notification sent to an external system (Slack, email, PagerDuty)

See also Human-in-the-Loop for StellarGate’s pre-LLM approval flow, which is a specific case of this general mechanism.

Real-time collaboration

Multiple users can work on the same chat thread, the same workflow run, the same document review in real time. Presence indicators show who’s currently on the screen. Cursor positions for text edits (in chat drafts, agent prompts) are shared live.

Cross-organization sharing

You can invite users outside your organization with tightly scoped access — typical use cases:

• External counsel reviewing a subset of a matter’s Base
• Clients reviewing the final deliverable (policy analysis, oversight report)
• Auditors inspecting audit logs without being able to read source documents
• Partner organizations collaborating on a joint research dataset

External users authenticate via SSO (SAML/OIDC) from their own identity provider, or via email + verification code. Their access is always scoped to the specific resource they’re invited to, with a configurable expiry date.

Notifications

Configure per-user notification preferences:

• In-app — bell icon, notification center
• Email — digest or immediate
• Slack / Teams — via connector
• Webhook — for integration with custom systems

Workflows can also fire notifications directly — “when a HIGH flag is created, page the duty lawyer,” “when a workflow completes, notify the reviewer channel.”

Audit log

Every collaborative action is logged immutably:

• Who viewed which document, when
• Who modified an agent prompt, with the before/after diff
• Who approved or rejected a workflow step
• Every comment, every mention, every @ invocation

The audit log is exportable in standard formats (JSON, CSV) for integration with SIEM tools. See Audit Log.

Version control

Agent prompts, workflow definitions, and custom rule sets are versioned. Every change is attributed and reversible. For workflows, you can compare two versions side-by-side and see which outputs would differ — useful when refining a reviewer’s playbook without risking regressions on historical data.

Related

• Auth & RBAC — role model in detail
• Audit Log — exportable audit trail
• Chat — conversational collaboration
• Workflows — approval gates in pipelines