Confirmed coverage
Integrations for the tools you actually operate
Agnys integration coverage currently includes Claude, LangChain, VS Code, LangGraph, AutoGen Studio, LM Studio, Ollama and Antigravity. These products occupy different layers of an agent stack: some are assistants or development surfaces, some orchestrate multi-step applications, and others run models locally. Agnys is designed to preserve a consistent evidence model across those differences.
Integration does not mean pretending every environment exposes identical telemetry. The useful capture point depends on how a tool runs and what an organization controls. Agnys can combine forwarder-based observation with SDK instrumentation, proxy capture, log collection and supported event interfaces. Teams should validate the chosen capture path against their versions, deployment architecture and evidence requirements before production rollout.
- Claude and Antigravity agent activity
- LangChain and LangGraph application workflows
- VS Code and AutoGen Studio development environments
- LM Studio and Ollama local model runtimes
Normalized evidence
Keep one operational language across different tools
Raw events from different products rarely use the same names, identifiers or level of detail. An orchestration framework may expose a graph node and tool call, while a local runtime exposes a model request and response. A development environment adds file and command activity around those model events. Agnys joins supported signals into sessions with timestamps, actors, event types and relationships that can be searched and replayed together.
Normalization helps engineering, security and governance teams investigate the same run without learning every vendor console. It also makes mixed stacks easier to review. A LangGraph workflow can call a locally served Ollama model while development happens in VS Code; the resulting record should still answer which session ran, what tools were invoked, what changed and where a human intervened.
- Session and agent attribution
- Model calls, tool use and workflow transitions
- File, command, approval, cost and outcome signals when available
- Chronological replay and portable PDF, CSV or JSON exports
Frameworks
Trace custom agents built with LangChain and LangGraph
LangChain and LangGraph are commonly used to compose model calls, tools, state and multi-step control flow. For these applications, direct instrumentation can preserve application-specific identifiers and workflow context that a network request alone may not contain. Agnys can use that context to connect a run, its component events and its resulting side effects.
The objective is not to replace framework-native debugging. Native traces are valuable during development. Agnys adds an operational evidence layer intended to persist across tools and audiences, with integrity checks, behavioral monitoring, oversight records and exports that can be reviewed outside the builder team.
Local AI
Include LM Studio and Ollama in the evidence boundary
Local model runtimes can reduce dependence on hosted APIs and give teams more control over where inference occurs. They can also create an observability gap if governance processes only review cloud-provider dashboards. Agnys supports LM Studio and Ollama so locally served model activity can participate in the same operational history as the surrounding agent workflow.
Local execution does not automatically settle privacy, security or compliance questions. Teams still need to define what content is captured, where records are stored, who can access them and how long they are retained. Agnys provides the trace and evidence mechanisms; the organization remains responsible for configuration and policy decisions.
Developer environments
Connect model activity to real development side effects
Claude, VS Code, AutoGen Studio and Antigravity can participate in workflows where an agent reads repositories, edits files, executes commands or coordinates other components. A model response by itself is not a complete account of that work. The useful record connects the initiating request with actions and outcomes in the environment where they occurred.
Agnys is intended to help teams reconstruct those sequences and review unusual behavior. Capture coverage varies with the environment and permissions available, so production evaluation should include representative tool calls, file operations, failures, retries and approval paths rather than relying on a successful demonstration alone.
Rollout method
Validate coverage before relying on the record
Start with one representative workflow for each integration. Document the expected model calls, tools, files, commands, approvals and outcomes. Run the workflow successfully, deliberately trigger a failure and verify that Agnys preserves the events needed to explain both results. Then test identity attribution, timestamps, sequence integrity, search, replay and export.
Repeat that validation after material tool, model, framework or capture-agent upgrades. Integration support is an operational capability, not a permanent guarantee that every future version will expose identical data. A short regression checklist gives teams stronger evidence than an unsupported compatibility badge.
- Confirm the exact product and version in scope
- Map expected events before testing capture completeness
- Test success, failure, approval and recovery paths
- Record known gaps and assign an owner for revalidation