Frontier models are already smart. The interesting problem is making agents safe, auditable, and useful inside organizations that have rules, regulators, and customers.
AgentLayer is a platform company building the enterprise AI Agent Orchestration Platform for industries that were never built for the agentic world. One product, one engagement model, one recurring monthly contract. Forward-deployed engineering is in the price.
The foundational layer. Six platform-core capabilities, productized as the engine every deployment runs on.
| Capability | What it does |
|---|---|
| 🧠 Orchestration | Agent loop, tool dispatch, multi-step planning, lifecycle and state handoff |
| 🔐 Identity & Access | Principal-vs-agent action delegation, KYA (Know-Your-Agent), credential scoping |
| 📜 Compliance & Audit | Append-only action logs, attribution, replay, citations on retrieval, regulator-aware |
| 🔁 Workflow | Stateful multi-step workflows, human-in-the-loop checkpoints, approval routing |
| 🔭 Observability | Heartbeat health, task lineage, session state, agent-action traceability |
| 🛡️ Governance | Policy framework, sovereignty controls, jurisdictional grounding |
The enabling layer. The @agentlayer/* packages are open-source primitives extracted from production. Each one is a focused pattern proven against real regulated-industry workloads, then absorbed into the engine that powers every AgentLayer deployment. Engineering teams can engage these primitives directly. No contract, no signup, just code.
| Package | Pattern | Status |
|---|---|---|
@agentlayer/orchestration |
Agent loop, tool dispatch, planning, lifecycle, state handoff | 🛠️ In preparation |
@agentlayer/identity |
Principal-vs-agent delegation, KYA, scoped credentials | 🛠️ In preparation |
@agentlayer/audit |
Append-only logs, attribution, replay, citations | 🛠️ In preparation |
@agentlayer/observability |
Heartbeat health, task lineage, session state, traceability | 🛠️ In preparation |
@agentlayer/agent-kevin |
Canonical internal-team-agents reference. Knowledge pipeline, task orchestration, autonomous heartbeat, plugin dispatch | 🟢 Open source · In production |
@agentlayer/studio |
The operator surface. Activity feed, agents, tasks, timeline, audit, approvals | 🟡 Live prototype. Closed commercial |
Every customer engagement deploys across one or more of these orthogonal surfaces. All supervised through AgentLayer Studio.
|
Engineering, ops, finance, back-office teams operating at AI multiples. The |
Support, onboarding, account management. Audit-trailed, human-in-the-loop on consequential decisions. |
Autonomous agents transacting with the business. AgentPay is the leading instance, on a regulated payments rail. |
The application layer. Each module is a deployable, industry-tuned configuration on the same engine. Customer-facing brand, structurally the same platform.
| Module | Brand | Industry | Status |
|---|---|---|---|
agent-fintech |
💸 AgentPay | Payments, financial operations, fintech | 🟡 In progress |
agent-gov |
🏛️ AgentGov | Government, public-sector, RFP response | Engagement open |
agent-edu |
🎓 AgentEdu | Education ops, registrar, admissions, student services | Engagement open |
agent-health |
🩺 AgentHealth | Healthcare operations, clinician-augmenting | Engagement open |
agent-legal |
⚖️ AgentLegal | Legal work, citation-disciplined, lawyer-augmenting | Engagement open |
agent-stay |
🏨 AgentStay | Hospitality operations, operator-augmenting | Engagement open |
The operator surface of the platform. Live today. studio.agentlayer.one
One screen showing every agent in the organization. What they're doing, what's queued for human review, what's complete, how they hand work to one another. The control plane for non-engineers: operators, executives, ops leads, supervisors.
Activity feed · Agents · Kanban + List · Timeline · Audit trail · Approvals queue · Notifications · Light and dark themes · Multi-tenant context · Real-time streaming
Studio is shipped under the same Enterprise Platform Contract. The way the AWS Console is part of AWS.
type AgentLayerStack = {
models: ['Claude (Anthropic) · default', 'GPT, Gemini · fallback'];
runtime: ['Bun + TypeScript', 'LangGraph', 'Anthropic Agent SDK'];
agents: ['Tool use', 'Planners + routers', 'Sub-agent orchestration', 'Memory: short/episodic/semantic'];
retrieval: ['Hybrid (BM25 + dense)', 'pgvector', 're-ranking', 'citation-aware'];
evals: ['Golden sets', 'LLM-as-judge', 'trace diffing', 'cost + latency budgets'];
data: ['PostgreSQL', 'Redis'];
cloud: ['AWS', 'Kubernetes', 'multi-region'];
frontend: ['Next.js 16', 'React 19', 'Tailwind CSS 4', '@agentlayer/ui'];
payments: ['x402', 'ERC-8183', 'ERC-8004', 'EIP-7702', 'Polygon · Arbitrum'];
};- Orchestration over framework. Patterns that compose with the language and tooling teams already use. No lock-in.
- Audit trail by default. Every agent action attributable, replayable, reviewable.
- Observability first. Heartbeat health, task lineage, session state. All surfaced.
- Human-in-the-loop is a feature. Designed-in approval points, not bolted-on guardrails. A core capability, not a fallback.
- Autonomy is earned, not assumed. Agents prove themselves through evals before they earn longer leashes.
| 🔗 Website | agentlayer.one |
| 🎛️ Studio (live) | studio.agentlayer.one |
| 🛠️ Developers | agentlayer.one/dev |
| 🏢 Engage | agentlayer.one/engage |
| 💬 Contact | contact@agentlayer.one |
Design agents. Govern agents. Trust agents.
© AgentLayer