Access Management

Where agents ask permission

Observe and permission across all your people and agents

Read the docs
One permissioned foundation

Enable permissionless building on a permissioned foundation

Connect your data and tools once. Honor source permissions and further scope down who can use them, what they can do, and when a person needs to step in.

Rules and approvals Source access Agent identities
Configurable source access lists

The right tools and data for all your people and agents.

Allowlist and blocklist tools and data down to the document level on top of source permissions. Give an agent, or human using an AI tool, only the documents it needs. Keep sensitive sources and actions out of reach.

Read the docs
Agent identity

Portfolio research agent

Read only

Deal room

Approved diligence documents
Allowed

Portfolio reports

Search and read
Allowed

Deal inbox

Read messages, no sending
Allowed

Restricted documents

Outside this agent’s scope
Blocked

Source permissions remain the upper limit.

Agent identities

An identity for every agent.

Mint unlimited agent identities with their own access controls and guardrails. Define who can assume each identity and the sources it can reach, without handing every agent all of a person’s permissions.

Read the docs
Observability

Know which AI touched which data.

Enable live logs with full agent traces to see who asked, which agent acted, the data and tools it used, and what was allowed or blocked. Follow each query through its tool calls and trace who canonicalized indexed values. Filter by person, agent, or access point, and export traces through OpenTelemetry to the observability tools you already use.

Read the docs
Source permissions, preserved

Honor source permissions.
Scope down for AI.

Honor the permissions from your connected systems. This includes in-place indexing as well as extracted structured data sets and fully indexed data with industry leading source permission propagation times. Keep AI access aligned as roles and group membership change, then narrow it further for each agent and workflow.

Data lineage

Source permissions follow every data point.

Extracted structured datasets inherit live-updating permissions from their source data. Trace each value all the way back to the document, line, or cell it came from. When source access, roles, or group membership changes, those permissions stay aligned across indexed data and derived datasets with industry leading source permission propagation times.

Read the docs
Rules and approvals

Lock down authorization with custom approval flows

Set custom authorization rules and approval flows for agents and people acting through AI tools. Require admin approval for sensitive actions, authentication challenges for human callers, and explicit approval windows so access is limited to the task at hand.

Read the docs
Approval workflow

A sensitive action needs a human

  1. 01

    Agent requests access

    Identity, source, and action are named.
  2. 02

    An approver reviews the request

    The agent waits for a decision.
  3. 03

    Approved access has an expiry

    Revoke it earlier whenever you need.
Explicit approval. Limited scope. Limited time.
Policy as code

Write it once. Version every change.

Craft and version repeatable, CEDAR-compatible policies. Define grants and organization-wide backstops, compare policy versions, and choose which version is active.

Read the docs

Access that expires

Set approval windows and revoke grants before they expire. Sensitive access does not have to become permanent access.

A record of who did what

Trace the person who initiated a call or canonicalized indexed data, the agent that made it, and the user or key it acted through.

A stop button for any agent

Deactivate an agent when its work is done or its access needs to change. The next call checks whether it is active.

Let your agents move fast.
Keep permission in your hands.

Talk to our team