Select an AI Agent integration method
The Integration Center connects AI Agents, Agent frameworks, self-built Agents, and their runtime behaviors to AgentLoop. After integration, you can use observability, audit, and evaluation capabilities based on the integration method you choose. Before you begin, select an integration method based on the Agent type, runtime environment, and target capabilities.
Integration method comparison
The following symbols are used in the comparison tables:
✅: Supported.
❌: Not supported.
Integration target coverage
Integration target | AI system runtime logs (eBPF) | LoongSuite Pilot | Language and framework probes |
Common Agents | ✅ Claude Code, OpenClaw, QwenPaw, Hermes, and more | ✅ Claude Code, Codex, Cursor, Qoder, Qoder CN, Qoder Work | ✅ OpenClaw, QwenPaw, Hermes, Coze |
Common Agent frameworks | ✅ LangChain, LangGraph, Dify, AgentScope, and more | ❌ | ✅ LangChain, LangGraph, AgentScope |
Python self-built Agents | ✅ | ❌ | ✅ |
Node.js self-built Agents | ✅ | ❌ | ✅ |
Rust self-built Agents | ❌ | ❌ | ❌ |
Other language self-built Agents | ✅ | ❌ | Partially supported or with defined capability limits. |
AI system runtime logs provide generic coverage for common Linux runtimes through eBPF. Codex uses Rust and a specialized TLS implementation, which limits LLM payload collection.
Language and framework probes require per-language, per-Agent, or per-framework adaptation.
Capabilities and runtime environments
Capability or environment
AI system runtime logs (eBPF)
LoongSuite Pilot
Language and framework probes
Observability
❌
✅
✅
Audit
✅
✅
❌
Evaluation
✅
✅
✅
Systems and environments
Linux hosts, Docker, Kubernetes
Linux, macOS, Windows development environments
Linux, macOS, Windows hosts and container environments
Upgrade impact
Manually upgrade the collector without restarting your workloads
Automatically upgrade collection components without restarting your workloads
Manually update on hosts, automatically update in containers. Requires application restart
The observability trace for AI system runtime logs is not yet available. eBPF may not directly obtain native Session, Turn, and Step identifiers or framework internal states.
The audit trace for language and framework probes is not yet available.
The capabilities listed above require that the integration target is supported and data collection succeeds.
Decision guidance
In supported Linux host or container environments, AI system runtime logs (eBPF) provide the broadest integration target coverage without code changes. However, eBPF does not currently support observability traces. Choose an integration method based on your scenario:
Need observability traces on macOS or Windows — Use language and framework probes.
Need observability traces for a supported AI Coding Agent — Use LoongSuite Pilot.
Need audit on Linux without code changes — Use AI system runtime logs (eBPF).
Need audit for a supported AI Coding Agent on macOS or Windows — Use LoongSuite Pilot.
Need the widest Agent coverage on Linux for evaluation — Use AI system runtime logs (eBPF).
For environments that require both runtime audit facts and native Agent context, you can combine multiple methods.
Find the corresponding integration item
Log on to the AgentLoop console. In the left-side navigation pane, click Integration Center. Select the corresponding integration item based on the integration target.
Integration target | Integration Center entry | Documentation |
AI Coding Agent | In AI Application Integration, select the corresponding item such as Claude Code, Codex, Cursor, Qoder, Qoder CN, or Qoder Work | |
Agents and platforms | In AI Application Integration, select the corresponding item such as OpenClaw, QwenPaw, Hermes, or Coze | Integrate OpenClaw applications, Integrate Hermes Agent applications |
Agent frameworks and self-built applications | In AI Application Integration, select the corresponding item such as AgentScope, LangChain / LangGraph, Dify, Python, or Node.js | Integrate AgentScope (Python) applications, Integrate Dify applications, Integrate LangChain & LangGraph applications |
AI system runtime | In AI Node Integration, select AI system runtime logs for Linux hosts, Docker, and Kubernetes environments |
To use AI Agent automated orchestration for application monitoring integration, see Integrate AgentLoop application monitoring through AI Agent.
Combined integration
AI system runtime logs, LoongSuite Pilot, and language and framework probes can be used in combination. These are not mutually exclusive integration methods:
AI system runtime logs collect generic model interactions and runtime facts through eBPF.
LoongSuite Pilot or language and framework probes supplement native Session, Turn, Step context, and framework states.
To simultaneously monitor Agent context alongside process, file, and network runtime behaviors, combine multiple integration methods.
Data masking
Integration configurations support on-demand client-side data masking for sensitive information such as phone numbers, national ID numbers, email addresses, IP addresses, landline numbers, and bank card numbers. Data masking is disabled by default. For specific configuration details, refer to the corresponding integration item.
What's next
Verify the integration status: Follow the specific integration documentation to verify that target data has been written and troubleshoot no-data issues.
View observable data: View model calls, token usage, Agent traces, and tool calling data.
Enable audit capabilities: Configure application, runtime risk detection, and semantic detection options in audit management. For more information, see Audit management.
Use audit facts: Browse sessions and view token analysis and behavior analysis. For more information, see Audit facts.
Run trajectory evaluation: Create trajectory datasets and configure evaluation tasks to analyze Agent execution quality.