Clasper Docs

Clasper

Production Agent Runtime with Governance & Observability

Clasper is an API-first, stateless agent execution platform designed for production SaaS integration. It enables workspace-driven agents with traceable behavior, governance controls, role-based access, cost/risk guards, and operational visibility — suitable for multi-tenant backends and audit-intensive environments.

“AI agents are not demos. They are production systems.”

Core Pillars

Pillar	Description
Agent Observability	Full execution traces with replay, diff, annotations & retention policies
Skill Runtime	Versioned YAML manifests with lifecycle states and testing
Governance & Safety	Tenant isolation, permissions, audit logs, redaction, risk scoring
Provider Abstraction	Normalized interface across LLM providers
Operational Tooling	Budget controls, cost forecasting, workspace pinning, environments, impact analysis

What Clasper Is

• Stateless HTTP runtime for agent executions
• Workspace-driven prompt config (SOUL.md, AGENTS.md, HEARTBEAT.md, skills)
• Multi-tenant context isolation with per-user scoping
• Skill registry with versioning, lifecycle states, and testing
• Observable and explainable traces with diff, replay, and annotations
• Operational guardrails (RBAC, budgets, risk scoring, audit logs)
• Control Plane Contract for portable backend integration
• Smart context selection (optional relevance-based skills + memory)

What Clasper Is Not

• A daemon for OS/browser automation (no shell access, no file system)
• A personal agent chatbot (designed for backend integration, not direct chat)
• A general automation framework (stateless, no persistent sessions)
• A replacement for your backend (your system remains the source of truth)

How Clasper Works With Your Backend

Clasper has a bidirectional relationship with your SaaS backend:

┌─────────────┐                              ┌─────────────┐
│   Your      │ ────── (1) send message ───▶ │   Clasper   │ ──▶ LLM
│   Backend   │                              │   Runtime   │
│             │ ◀── (2) agent calls APIs ─── │             │
└─────────────┘                              └─────────────┘
     │                                              │
     │  Source of truth:                            │  Stateless:
     │  • Users, auth                               │  • Loads workspace config
     │  • Tasks, messages                           │  • Builds prompts
     │  • Conversations                             │  • Routes LLM calls
     │  • Documents                                 │  • Mints agent JWTs
     └──────────────────────────────────────────────┘

1. Your backend sends messages to Clasper (POST /api/agents/send)
2. Clasper calls an LLM with workspace-configured prompts (personas, rules, skills)
3. The agent response may call your APIs to create tasks, post messages, etc.
4. Your backend remains the source of truth — Clasper is stateless

This means you can run multiple Clasper instances behind a load balancer with no sticky sessions.

Execution Modes

Mode	Endpoint	Description
Request/Response	POST /api/agents/send	Synchronous agent execution
Streaming (SSE)	POST /api/agents/stream	Real-time streaming responses
LLM Task	POST /llm-task	Structured JSON-only output
Trace Replay	GET /traces/:id/replay	Reproduce past executions
Ops Console	/ops/*	OIDC-protected operational UI

Multi-Provider LLM Support

Clasper supports multiple LLM providers out of the box:

Provider	Models
OpenAI	GPT-4o, GPT-4.1, etc.
Anthropic	Claude 4, Claude 3.5, etc.
Google	Gemini 2.5, Gemini 2.0, etc.
xAI	Grok 2, Grok 2 Mini
Groq	Llama 3.3, Mixtral
Mistral	Mistral Large, Codestral
OpenRouter	Multiple providers

Quick Links

• Getting Started — Run your first agent
• Architecture — How it works
• Integration Guide — Connect to your backend
• Governance — RBAC, budgets, and audit
• API Reference — Full endpoint documentation