Know which agent
did that.
Agents are a new class of non-human actor: they call tools, hit APIs and act continuously, often with nobody watching. Agent Identity gives each one a stable, traceable identity — stamped on every tool call, LLM turn and audit entry, and presentable to the resources it calls so they can verify the caller instead of trusting a name in a log.
from promptise.identity import AgentIdentity
agent = await build_agent(
model="anthropic:claude-sonnet-4-5",
servers={...},
identity=AgentIdentity.from_entra(
"billing-bot", resource="api://mcp",
),
)
# every action is attributed to billing-bot,
# and every call carries a signed credentialTwo tiers. Start with a name, upgrade to proof.
Identity is not the model’s API key. It answers a different question — who is acting — so it can be attributed, authorized and audited.
Local identity
A stable id, a display name, an owner and metadata, asserted by the process itself. Free, instant, and enough to answer “which agent did this?” across a fleet — in traces, audit entries, runtime journals and every tool call.
AgentIdentity("billing-bot", owner="payments")Verifiable identity
Backed by your platform’s identity provider, which mints a short-lived signed JWT. The agent presents it to the resources it calls, so they verify the caller instead of trusting a self-asserted id — and one identity can carry a credential per audience.
identity.get_credential("api://billing")Your platform already issues identities. Use them.
No new directory to run, no secrets to rotate yourself: the credential comes from the identity provider you already have — and AgentIdentity.auto() detects which one you are on.
AgentIdentity.from_entra()Identity source: IMDS, or the federated token file on AKS. The credential is short-lived and refreshed before it expires; nothing long-lived is stored by Promptise, and persistence stays in your IdP.
One line at build time. Everywhere after that.
Every LLM turn and tool call in the trace carries the agent id — filter a fleet’s activity by who acted.
A verifiable identity presents its signed credential to the servers it calls, scoped per resource.
Approval requests and HMAC-chained audit entries name the agent, not an anonymous process.
Runtime journals, escalations and cross-agent messages all carry it, so a fleet stays attributable.
Note: this page is general technical information, not legal, regulatory or compliance advice. Using Promptise does not by itself make you or your product compliant with any law or standard — consult a qualified professional before relying on anything here.