Metatextai
for LuumenAI
Inspect Metatext AI guardrails, then change them as an approved step
Connect Metatext AI and the agent can read your applications, models, and guardrail policies while you work. Ask what a policy covers today, classify or extract from sample text, or check which red team probes exist. When a policy needs to change, the agent shows the edit first and waits for your approval before sending it. Authentication is by API key.
The Metatextai toolbox
13 tools: 9 read, 4 write. Reads answer instantly. Writes require approval by default. Everything is logged.
- ReadChat CompletionsGenerate chat completions.
- ReadClassify TextClassify text.
- ReadEvaluate MessagesEvaluate LLM messages against policies/guardrails.
- ReadRun ExtractionRun information extraction.
- ReadGenerate TextGenerate text for a project model.
- ReadList ApplicationsRetrieve a list of all existing applications.
- ReadList ModelsRetrieve a list of all available models and their supported tasks.
- ReadList Guardrail PoliciesList all guardrail policies for a specific application.
- ReadList Red Team Test ProbesList all available red team test probes.
- WriteCreate Policy GuardrailsCreate a policy guardrail. Approval by default
- WriteDelete Guardrail PolicyDelete a guardrail policy. Approval by default
- WriteRun Red Team Test ScanRun a vulnerability red-team test scan. Approval by default
- WriteUpdate Policy GuardrailsUpdate an existing policy's guardrails. Approval by default
One prompt, start to finish
What a governed Metatextai run looks like inside Luumen.
Questions
Authorize once with API token. Luumen lists the scopes each action needs before you approve the connection, and credentials never appear in the chat.
Read actions answer immediately. Anything that writes — create policy guardrails, delete guardrail policy, run red team test scan, update policy guardrails — is shown as a plan and requires approval by default, including the 1 action classified as destructive. Administrators configure that per tool, so you decide exactly which actions can ever run unattended.
You decide. Actions are granted per agent, skill, and team, and per environment — production is not staging. Read access can be broad while writes stay narrow.
Every call to Metatextai — read or write, approved or declined — is recorded with the actor, the input, and the result, and can be linked to the ticket or change record.
Often connected alongside
All integrations
Adobe Marketing Agent MCPAdobe Experience Platform context, sandbox switches, and async task status
AffindaRead parsed documents, collections, and annotations — change them with approval
Agent MailGive the agent its own mailbox for reading threads and sending updates
AI/ML APIInspect assistants, threads, and runs — and change them as approved stepsPut Metatextai to work with Luumen
Connect in minutes. Every action scoped, approved, and audited from day one.