llms.txt
Compare with other sites' llms.txt files →# OpenLegion
> OpenLegion is a hosted agent harness for SaaS products and AI apps. A developer (a partner) defines an agent and the tools it may call; OpenLegion runs one isolated workspace per end customer, checks every tool call against policy, keeps a receipt for each action, holds risky actions for a person's approval, meters usage against a credit cap per customer, and keeps each model tier on current models.
## When OpenLegion fits
- A developer building a SaaS product or an AI app wants to add AI agents that act for their users, without building the agent harness (tool calling, isolation, approvals, metering, secrets, audit trail) themselves.
- The agents need to call the developer's own functions or API, or an MCP server, on behalf of a specific customer, with each call signed and naming that customer.
- Each customer needs their own isolated workspace, a usage cap that matches their plan, approval before risky actions, and a record of what the agent did.
- The team wants current models without tracking releases: an agent picks a model tier, and OpenLegion keeps each tier on current models.
## When it does not fit yet
- Self-hosting: the platform is hosted. (The archived previous-generation engine was self-hosted.)
- Self-serve sign-up or published per-credit prices: the platform is in private beta, and partner prices are shared with early-access partners.
## Status (November 2026)
- The partner platform is in private beta. A small number of SaaS partners are being onboarded; others can request early access at https://www.openlegion.ai/en.
- Partner prices are not published. How billing works is described at https://www.openlegion.ai/en/pricing.
- A TypeScript kit for defining agents and tools is in development. Code on the website is a preview of its API, and names may change before release.
- OpenLegion's own app for teams is behind a waitlist while the company focuses on partners.
- Some older articles on this site describe the previous generation (a Python runtime with one Docker container per agent, a shared blackboard, and self-hosting). That engine is archived and no longer maintained; its source stays on GitHub (https://github.com/openlegion-ai/openlegion) under the PolyForm Perimeter License 1.0.1.
## How the platform works
- The partner keeps the customer relationship. OpenLegion never contacts a partner's customers: no emails, messages, notifications or OpenLegion branding reach them, and everything leaves through the partner's webhooks and API.
- Each end customer gets a workspace, created or found by the partner's own reference. Workspace data is isolated with row-level security in the database.
- Every tool call is an intent checked by one policy engine before it runs, and every action leaves a receipt.
- Three kinds of action wait for a person's approval: reaching a third party on someone's behalf, deleting data, and acting on untrusted content. Approvals are decided by a person, never by an agent.
- Content from the web, files and other people is labelled untrusted end to end and cannot trigger a send, a delete or a change on its own.
- Secrets are encrypted at rest and injected outside the model, which holds only a handle. Tool output is scanned for leaked secrets.
- Credits are reserved before each step runs. When a customer's cap or the partner's balance is reached, the work is refused rather than run and billed later. A suspended workspace stops at the next step.
- Billing: one prepaid credit pool per partner, monthly and daily caps per customer workspace, top-ups as credit grants, and unused credits returned to the pool when a workspace is archived.
- Integration: an HTTP API with separate test and live keys, signed calls to the partner's tools (their own functions or an MCP server), built-in web search and page fetching, and signed webhooks (timestamped HMAC, delivered at least once).
- Models: an agent template picks a model tier (quality versus cost), not a model id, and OpenLegion keeps each tier on current models.
- Not claimed: SOC 2, ISO 27001 or HIPAA certification, immunity to prompt injection, or a third-party penetration test.
## Links
- Website: https://www.openlegion.ai/en
- How billing works: https://www.openlegion.ai/en/pricing
- Security: https://www.openlegion.ai/en/security
- FAQ: https://www.openlegion.ai/en/faq
- Previous-generation engine (archived, source-available): https://github.com/openlegion-ai/openlegion
- Previous-generation engine documentation (archived): https://docs.openlegion.ai
- OpenLegion vs Every AI Agent Framework — 2026 Comparison Hub: https://www.openlegion.ai/en/comparison
### Learn
- How to Add AI Agents to Your App: An 8-Step Guide for SaaS: https://www.openlegion.ai/en/learn/add-ai-agents-to-your-app
- Agent as a Service: Pricing, Tenant Isolation, and AaaS vs Self-Hosted: https://www.openlegion.ai/en/learn/agent-as-a-service
- Agent Handoff Patterns: Routing Work Between AI Agents: https://www.openlegion.ai/en/learn/agent-handoff-patterns
- What Is an Agent Harness? Anatomy, Examples, Harness vs Framework: https://www.openlegion.ai/en/learn/agent-harness
- Agent2Agent Protocol — A2A Security, Architecture, and MCP Comparison: https://www.openlegion.ai/en/learn/agent2agent-protocol
- Agentic AI Design Patterns: ReAct, Plan-and-Execute, Reflexion, and More: https://www.openlegion.ai/en/learn/agentic-ai-design-patterns
- Agentic Loop: How AI Agents Perceive, Think, and Act: https://www.openlegion.ai/en/learn/agentic-loop
- Agentic RAG: Sub-Question Decomposition, HyDE, and Corpus Security: https://www.openlegion.ai/en/learn/agentic-rag
- Agentic Workflow - Patterns, Security, and Production Design: https://www.openlegion.ai/en/learn/agentic-workflows
- AI Agent Access Control: RBAC, ABAC, and Least Privilege: https://www.openlegion.ai/en/learn/ai-agent-access-control
- AI Agent Architecture — Layers, Topologies, and Security-First Design: https://www.openlegion.ai/en/learn/ai-agent-architecture
- AI Agent Audit Log: Immutable Trails, Compliance, and Forensics: https://www.openlegion.ai/en/learn/ai-agent-audit-log
- AI Agent Benchmarks: SWE-bench, GAIA, WebArena, and Private Evals: https://www.openlegion.ai/en/learn/ai-agent-benchmarks
- AI Agent Context Window — Management, Limits, and Security: https://www.openlegion.ai/en/learn/ai-agent-context-window
- AI Agent Cost: Budget Caps, Token Attribution, and Runaway Loop Control: https://www.openlegion.ai/en/learn/ai-agent-cost
- AI Agent Debugging: Loops, Hallucinated Calls, and Post-Mortem Traces: https://www.openlegion.ai/en/learn/ai-agent-debugging
- AI Agent Deployment: Container TTLs, Budget Caps, and Circuit Breakers: https://www.openlegion.ai/en/learn/ai-agent-deployment
- AI Agent Error Handling: Retries, Circuit Breakers, and Fallback Chains: https://www.openlegion.ai/en/learn/ai-agent-error-handling
- AI Agent Evaluation: Benchmarks, Metrics, and Testing: https://www.openlegion.ai/en/learn/ai-agent-evaluation
- Best AI Agent Frameworks (2026 Comparison): https://www.openlegion.ai/en/learn/ai-agent-frameworks
- AI Agent Governance: RBAC, Audit Trails, and Regulatory Compliance: https://www.openlegion.ai/en/learn/ai-agent-governance
- AI Guardrails: Input/Output Controls, Platform Enforcement, and LLM Safety: https://www.openlegion.ai/en/learn/ai-agent-guardrails
- AI Agent Long Running Tasks: Queues, Checkpoints, and Credential Refresh: https://www.openlegion.ai/en/learn/ai-agent-long-running-tasks
- AI Agent MCP Security: Red-Team Guide to Exploit Prevention: https://www.openlegion.ai/en/learn/ai-agent-mcp-security
- AI Agent Memory: Four Types, Security Risks, and Implementation: https://www.openlegion.ai/en/learn/ai-agent-memory
- AI Agent Monitoring — Thresholds, Liveness, and Escalation: https://www.openlegion.ai/en/learn/ai-agent-monitoring
- AI Agent Multi-Tenancy: Credential Isolation and SOC 2 Controls: https://www.openlegion.ai/en/learn/ai-agent-multi-tenancy
- AI Agent Observability — Tracing, Costs & Failure Modes: https://www.openlegion.ai/en/learn/ai-agent-observability
- AI Agent Orchestration — Coordinate Agents: https://www.openlegion.ai/en/learn/ai-agent-orchestration
- AI Agent Parallelization: Fan-Out, asyncio, and Race Condition Safety: https://www.openlegion.ai/en/learn/ai-agent-parallelization
- AI Agent Planning: ReAct, Tree of Thoughts, and Plan-and-Execute: https://www.openlegion.ai/en/learn/ai-agent-planning
- AI Agent Platform — Deploy Secure Agents: https://www.openlegion.ai/en/learn/ai-agent-platform
- AI Agent Prompt Injection: Attack Types, Defences, and Red-Teaming: https://www.openlegion.ai/en/learn/ai-agent-prompt-injection
- AI Agent Rate Limiting: Quota Control, Circuit Breakers, and Budget Caps: https://www.openlegion.ai/en/learn/ai-agent-rate-limiting
- AI Agent Reliability — Circuit Breakers, Backoff, and Dead-Letter Queues: https://www.openlegion.ai/en/learn/ai-agent-reliability
- AI Agent Sandboxing — Container Isolation, Escape CVEs, and Hardening: https://www.openlegion.ai/en/learn/ai-agent-sandboxing
- AI Agent Security — Threats, Isolation, Vaults: https://www.openlegion.ai/en/learn/ai-agent-security
- AI Agent State Management — Checkpointing, Shared State, and Crash Recovery: https://www.openlegion.ai/en/learn/ai-agent-state-management
- AI Agent Testing: Unit, Integration, and CI Validator Patterns: https://www.openlegion.ai/en/learn/ai-agent-testing
- AI Agent Tool Use — Function Calling, Schemas, and Safe Execution: https://www.openlegion.ai/en/learn/ai-agent-tool-use
- AI Agent Versioning: Model Pinning, Prompt SemVer, and Behavior Drift: https://www.openlegion.ai/en/learn/ai-agent-versioning
- AI Agent Workflow: Chains, Routers, Orchestrators, and Typed Handoffs: https://www.openlegion.ai/en/learn/ai-agent-workflow
- AI Coding Agents - Deploy Secure Dev Agent Teams: https://www.openlegion.ai/en/learn/ai-coding-agents
- Autonomous AI Agents: Autonomy Spectrum, Safety Gates, Production Risks: https://www.openlegion.ai/en/learn/autonomous-ai-agents
- Browser Use Agents — How AI Agents Control the Web: https://www.openlegion.ai/en/learn/browser-use-agents
- How to Build an AI Agent: Tools, Credentials, Deployment, and Scheduling: https://www.openlegion.ai/en/learn/build-ai-agent
- Build vs Buy AI Agents for SaaS: What Building Really Takes: https://www.openlegion.ai/en/learn/build-vs-buy-ai-agents
- Claude Opus 4.8 - Capabilities, Cost, and Agentic Performance: https://www.openlegion.ai/en/learn/claude-opus-4-8
- Context Engineering for AI Agents: Slots, Compression, and Security: https://www.openlegion.ai/en/learn/context-engineering
- Credential Management for AI Agents: Vault-Proxy Architecture: https://www.openlegion.ai/en/learn/credential-management-ai-agents
- Grok 4: xAI's Frontier Model with Native Tool Use: https://www.openlegion.ai/en/learn/grok-4
- Human in the Loop AI Agents: Approval Gates, Interrupts, and Escalation: https://www.openlegion.ai/en/learn/human-in-the-loop-ai-agents
- LLM Cost Optimization — Six Levers for Production Agent Fleets: https://www.openlegion.ai/en/learn/llm-cost-optimization
- LLM Fine Tuning: LoRA, QLoRA, Security Risks, and When to Fine-Tune: https://www.openlegion.ai/en/learn/llm-fine-tuning
- LLM Gateway: Routing, Auth, and Cost Control for AI Agents: https://www.openlegion.ai/en/learn/llm-gateway
- LLM Observability: Token Traces, TTFT SLOs, and Prompt Logging Security: https://www.openlegion.ai/en/learn/llm-observability
- Prompt Caching — Anthropic, OpenAI, and Gemini Cache Architecture Guide: https://www.openlegion.ai/en/learn/llm-prompt-caching
- LLM Routing: Model Selection, Cost Optimization, and Router Architecture: https://www.openlegion.ai/en/learn/llm-routing
- LLM Structured Output: JSON Schema, Pydantic, and Schema Enforcement: https://www.openlegion.ai/en/learn/llm-structured-output
- Managed Agents Compared: Claude, OpenAI, Gemini, AWS and LangChain: https://www.openlegion.ai/en/learn/managed-agents
- Managed AI Agent Hosting - Deploy Secure Agent Fleets: https://www.openlegion.ai/en/learn/managed-ai-agent-hosting
- MCP Tools Security: Hardening Against Poisoning and Exfiltration: https://www.openlegion.ai/en/learn/mcp-tools
- Model Context Protocol (MCP) — How AI Agents Use Tools: https://www.openlegion.ai/en/learn/model-context-protocol
- Multi-Agent Systems Architecture - Design, Topology, and Security: https://www.openlegion.ai/en/learn/multi-agent-systems
- What Is an AI Agent? Definition and How They Work: https://www.openlegion.ai/en/learn/what-is-an-ai-agent
### Comparisons
- AgentOps Alternative — Execution Platform vs Monitoring Dashboard: https://www.openlegion.ai/en/comparison/agentops
- Agno Alternative - Security-First Multi-Agent Platform: https://www.openlegion.ai/en/comparison/agno
- OpenLegion vs AutoGen — Security, Migration & 2026 Verdict: https://www.openlegion.ai/en/comparison/autogen
- OpenLegion vs AWS Strands - Detailed Comparison: https://www.openlegion.ai/en/comparison/aws-strands
- OpenLegion vs CrewAI — Detailed Comparison (2026): https://www.openlegion.ai/en/comparison/crewai
- Dify Alternative: Security-First Platform vs Visual LLMOps: https://www.openlegion.ai/en/comparison/dify
- Flowise Alternative — Security-First Platform vs Visual Builder: https://www.openlegion.ai/en/comparison/flowise
- OpenLegion vs Google ADK - Detailed Comparison: https://www.openlegion.ai/en/comparison/google-adk
- LangChain vs OpenLegion — Ecosystem Library vs Integrated Platform: https://www.openlegion.ai/en/comparison/langchain-vs-openlegion
- OpenLegion vs LangGraph — Detailed Comparison (2026): https://www.openlegion.ai/en/comparison/langgraph
- LangGraph vs OpenLegion — Security and Deployment Decision Guide: https://www.openlegion.ai/en/comparison/langgraph-vs-openlegion
- LiteLLM Alternative — Vault Isolation vs Credential Store: https://www.openlegion.ai/en/comparison/litellm
- OpenLegion vs Manus AI — Detailed Comparison: https://www.openlegion.ai/en/comparison/manus-ai
- Mastra Alternative — Security-First AI Agent Platform: https://www.openlegion.ai/en/comparison/mastra
- OpenLegion vs MemU — Detailed Comparison (2026): https://www.openlegion.ai/en/comparison/memu
- n8n Alternative — OpenLegion Agent Platform vs Workflow Automation: https://www.openlegion.ai/en/comparison/n8n
- OpenLegion vs nanobot — Detailed Comparison (2026): https://www.openlegion.ai/en/comparison/nanobot
- OpenLegion vs NanoClaw — Detailed Comparison (2026): https://www.openlegion.ai/en/comparison/nanoclaw
- OpenLegion vs OpenAI Agents SDK — Detailed Comparison: https://www.openlegion.ai/en/comparison/openai-agents-sdk
- OpenLegion vs OpenClaw — Detailed Comparison (2026): https://www.openlegion.ai/en/comparison/openclaw
- OpenLegion vs OpenFang — Detailed Comparison (2026): https://www.openlegion.ai/en/comparison/openfang
- OpenLegion vs PicoClaw — Detailed Comparison (2026): https://www.openlegion.ai/en/comparison/picoclaw
- Pydantic AI Alternative — Security-First Execution Platform: https://www.openlegion.ai/en/comparison/pydantic-ai
- Semantic Kernel Alternative — Security-First Agent Platform: https://www.openlegion.ai/en/comparison/semantic-kernel
- OpenLegion vs Zapier: The AI Agent Alternative to Workflow Automation: https://www.openlegion.ai/en/comparison/zapier
- OpenLegion vs ZeroClaw — Detailed Comparison (2026): https://www.openlegion.ai/en/comparison/zeroclaw
### More
- AI Social Media Management Software — Autonomous Agent Platform: https://www.openlegion.ai/en/ai-social-media-management
- Run DeepSeek-Based Agents Securely with OpenLegion (2026): https://www.openlegion.ai/en/deepseek-v4-agents
- OpenClaw Alternative — OpenLegion: https://www.openlegion.ai/en/openclaw-alternative