About Record

Software was built for people. The next decade is built for agents.

Record exists so enterprises can let agents do real work without losing ownership of identity, access, credentials, policy, approvals, enterprise knowledge, incidents, diagnostics, spend, or proof. We call that ITSM for AI agents.

Start a 4-week pilotFor developers

Agents are moving from chat to work. Work requires access. Access requires accountability. Record is the system where enterprises put agents, credentials, approvals, policies, knowledge, actions, incidents, diagnostics, and audit evidence on the record.

Why ITSM for AI agents

Enterprises already know how to govern work. The worker changed.

The enterprise will not split into two worlds. For years, humans and agents will run the same systems together — the same IdP, CRM, ticket queue, codebase, and APIs. Traditional ITSM already gives human work a system: requests, approvals, access, incidents, knowledge, spend, and audit. Agents now do that same work at machine speed, so they need the same operating model — with stronger enforcement and clearer evidence.

Human ITSM manages
Agent work now needs
01Service requests and fulfillment
Agents request models, tools, skills, APIs, data, and enterprise actions through governed paths.
02Identity and access management
Every agent acts as a known principal, on behalf of a named human, with scoped access and no raw credential custody.
03Approvals and change governance
Risky actions pause for policy-driven human review, then resume with approver, reason, timestamp, and evidence.
04Incident and problem management
Signals become owned Issues with diagnostics, root cause, impact, lifecycle, and evidence for resolution.
05Knowledge base articles and runbooks
Record Brain gives agents enterprise knowledge, memory, prior decisions, approvals, incidents, and patterns.
06Spend, audit, and accountability
Every model call, tool call, credential lease, approval, incident, and dollar is attributed and queryable.

What changed

Agents turn automation into a managed workforce.

Agents are becoming a workforce

They read systems, call tools, use credentials, escalate work, and operate when humans are not watching. That is not a chatbot problem. It is an operating model problem.

Enterprise systems were built for people

IAM, ITSM, SIEM, and ticket queues assume a human subject. Agents need identity, access, policy, approvals, spend controls, observability, and lifecycle built around machine-speed work.

Access is the moment of truth

The risky moment is not when an agent generates text. It is when it touches a system, uses a credential, moves data, changes state, or triggers a workflow with real consequences.

Evidence is a product feature

The enterprise needs to answer which agent acted, on whose behalf, under which policy, with which credential, what failed, who owns the issue, and what evidence proves the outcome.

What we believe

Autonomy only works when authority is managed.

01No agent action should be anonymous.
02No agent should hold standing raw credentials.
03No important action should happen without policy context.
04No risky workflow should depend on a prompt behaving correctly.
05No audit question should require stitching four systems together.

The platform

Two layers. One control plane.

Any agent connects through Record Gateway for identity, tools, models, policy, access, and audit. Agents Record runs add a deeper runtime — sandbox lifecycle, durable approvals, and kernel-level enforcement — under the same policy and the same record.

Any agent connects

Record Gateway

One governed route between any agent — local, ADK, LangChain, custom — and your enterprise systems.

Identity & access
Every agent acts as a known principal, on behalf of a named human, with scoped access and no raw-credential custody.
Model gateway
100+ models behind one route, with selection, fallback, content filters, rate limits, and per-agent spend.
Tool & capability registry
MCP servers, OpenAPI specs, and agent-to-agent endpoints federated into governed capabilities, evaluated per call.
Policy & approvals
One Cedar policy plane with durable human-in-the-loop review on risky actions.
JIT access & credentials
Short-lived, scoped credentials resolved at Record edges. Agents receive capabilities and results, never raw secrets.
Knowledge & memory
Record Brain gives agents enterprise knowledge, memory, prior decisions, and operational cases that compound each run.
Budget & spend
Hard caps before the burn, with every dollar attributed to a user, team, and agent — so finance can explain the model bill and control it.
Observability & audit
Traces, policy decisions, credential leases, approvals, and incidents in one queryable record.
Agents Record runs — deeper

Record Runtime

When the workflow needs more than a governed route, Record runs the agent and owns its execution end to end.

Sandbox lifecycle & warm pools
Managed pod lifecycle with pre-warmed pools for sub-second session adoption.
Kernel-level enforcement
A sandbox sidecar enforces policy at the syscall and network layer, injecting credentials at egress.
Durable HITL resume
Risky work pauses durably and resumes with approver, reason, timestamp, and policy evidence.
Runtime state & resume
Runtime state survives restarts, so long-running agents resume exactly where they left off.

The old question was: who clicked the button?

The agent-era question is sharper — which agent acted, on whose behalf, under which policy, with which credential, and what evidence proves it? Record exists so the enterprise can answer that before the regulator, the board, or the incident team asks. Agents get the capability to work; Record keeps the credentials, the policy, and the proof.