Skip to content
@AgentLayer1

AgentLayer

The platform agents run on.

AgentLayer

The platform agents run on.

Industries built for agents. Forward-deployed engineering.


agentlayer.one studio (live) developers agent-kevin contact



Frontier models are already smart. The interesting problem is making agents safe, auditable, and useful inside organizations that have rules, regulators, and customers.


🧭 About

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 Engine

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

🧩 Agentic Primitives

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
Closed patterns stay proprietary. Open primitives compose.

🚀 Deployment Patterns

Every customer engagement deploys across one or more of these orthogonal surfaces. All supervised through AgentLayer Studio.

🧑‍💻 Internal-team agents

Engineering, ops, finance, back-office teams operating at AI multiples. The @agentlayer/agent-kevin pattern, configured for the customer's organization.

📞 Customer-facing agents

Support, onboarding, account management. Audit-trailed, human-in-the-loop on consequential decisions.

🤝 Agent-as-customer

Autonomous agents transacting with the business. AgentPay is the leading instance, on a regulated payments rail.


🏭 Industry Modules

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

🎛️ AgentLayer Studio

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.


⚒️ Stack

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'];
};

🧱 Engineering Principles

  • 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.

🛰️ Connect

🔗 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

Popular repositories Loading

  1. agent-kevin agent-kevin Public

    Portable, file-based personal AI assistant.

    TypeScript 2 1

  2. vscode-cmux-open vscode-cmux-open Public

    Open any folder as a cmux workspace from the VS Code Explorer context menu.

    TypeScript 1

  3. agentlayer-claude-marketplace agentlayer-claude-marketplace Public

    Agentic primitives, identity-aware assistants, and industry-tuned skill packs

    1

  4. .github .github Public

Repositories

Showing 4 of 4 repositories

People

This organization has no public members. You must be a member to see who’s a part of this organization.

Top languages

Loading…

Most used topics

Loading…