Outdated Drivers Are Slowing You Down
One free scan finds every outdated or missing driver and matches the right update for your exact hardware.Free scan · exact hardware matchPC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11OpenTelemetry’s GenAI semantic conventions give agent developers a shared way to describe agent invocations, model inference, planning, tools, and related metrics in telemetry. As of October 4, 2026, the documentation index marks these conventions Development, so treat them as evolving guidance and verify the current specification and your language’s implementation support before adopting them.
What the conventions standardize
Semantic conventions define names and meanings for telemetry so traces and metrics can describe comparable operations. The OpenTelemetry GenAI conventions cover spans, metrics, events, exceptions, inference-token metrics, MCP, and provider-specific patterns. Their human-readable pages are generated substantially from YAML models; the repository also maintains reference implementations and tooling. OpenTelemetry GenAI semantic-conventions repository.
As an Amazon Associate I earn from qualifying purchases.
They are not a requirement that every application capture every field, nor do they guarantee that every language’s instrumentation implements every signal. The specification distinguishes recommended patterns from fields that are conditional, opt-in, or dependent on what the instrumentation can identify.
How to model an agent invocation
Keep the agent’s higher-level invocation distinct from model inference and tool execution. That separation lets a trace show what the agent did without collapsing its work into one model-call span.
#1 Best Overall
Remote and in-process agents
For a remote agent, the convention describes a client span. For an agent invoked in the same process, it describes an internal span. In both cases, use invoke_agent as the operation name. When the agent name is readily available, the suggested span name is invoke_agent {gen_ai.agent.name}; otherwise use invoke_agent. These are recommended patterns, not a mandate to invent unavailable identity data. Agent span conventions.
Agent creation and identity
Creation is a separate operation: a remote agent creation is represented with a create_agent client span and suggested span name. Record gen_ai.agent.name for a human-readable name, gen_ai.agent.id for a stable unique identifier where applicable, and gen_ai.agent.version when available. Do not substitute a transient in-memory instance identifier for the identity of a hosted agent.
gen_ai.system_instructions is explicitly opt-in. Its presence in the schema does not mean system instructions should be captured by default.
Illustrative trace tree
This sketch illustrates the relationships described by the conventions; it is not a promise that every framework emits this exact tree:
invoke_agent (client, if remote; internal, if in-process)
├── plan (only if planning is distinguishable)
│ └── model inference
├── tool execution
└── model inferenceA model call used for planning is a child of the plan span. Resulting tool or task spans are typically siblings of the plan span beneath the invocation. Use the structure to preserve actual operation relationships rather than adding spans to make a trace look more elaborate.
When to add planning, workflow, and tool spans
Planning
Use a plan span for an identifiable planning or task-decomposition operation. The convention says not to emit one when instrumentation cannot distinguish planning from generic reasoning or ordinary inference. A model call alone is not evidence that the application performed a separately identifiable planning operation.
Workflows
A workflow invocation can be represented by an internal invoke_workflow span, with the workflow name in the suggested span name when it is available. The convention gives examples spanning graph- and crew-style execution.
Recommended Free Tools
Tools and sub-agents
Represent tool execution in the agent’s call tree and record success or error according to OpenTelemetry’s error-recording guidance. Distinguish client-side tools run by the agent or framework from tools executed internally by a model provider. That distinction affects both the trace and the tool-call metric.
When an agent delegates to a sub-agent, the sub-agent’s work belongs to that agent’s own invocation. Preserve the invocation boundary instead of attributing all nested activity to the parent as if it were one agent’s direct work.
How provider identity works
gen_ai.provider.name identifies the provider-specific telemetry flavor used by the instrumentation. Set it according to the instrumentation’s best knowledge and keep it aligned with relevant provider-specific attributes and signals. The best-known provider may be a proxy or hosting platform rather than the company that created the upstream model. GenAI span conventions.
Generic GenAI client spans describe logical operations such as inference, embeddings, retrieval, fetch response, and memory. A span should cover the logical operation through receipt of the full response, or until it ends because of an error or cancellation; automatic retries belong within that logical span. Where a provider-specific convention exists, treat it as an extension or override of generic guidance rather than assuming identical attributes across providers.
What’s actually slowing this PC down?
Pick the symptom - the matching free tool is one click away.
The documentation index lists conventions for Anthropic, Azure AI Inference, AWS Bedrock, and OpenAI, and links MCP conventions separately. Check the current index and applicable provider page when choosing attributes: GenAI conventions index.
Best Value
What the agent metrics count
The documented agent metrics are gen_ai.invoke_agent.duration, gen_ai.invoke_agent.inference_calls, and gen_ai.invoke_agent.tool_calls. The conventions recommend recording them alongside the relevant internal invocation span when applicable. Agent metric conventions.
- Scope: counts are scoped to an invocation. A sub-agent’s calls belong to that sub-agent’s invocation.
- Failed calls: include failed calls as specified by the conventions; do not silently compare counts that use different failure treatment.
- Tool attribution: count calls issued by the agent itself and avoid counting one tool call twice across the call tree.
- Provider-side tools: tools executed on the provider’s server, such as built-in search or code execution, are outside the client-side tool-call metric.
When comparing implementations, check whether each uses the same remote-client or internal invocation pattern, exposes agent identity and version, can reliably identify planning, distinguishes client-side from provider-side tools, and supports the relevant metrics in its language. These are useful scope checks, not a ranking of vendors or frameworks.
Events and content capture are not universal defaults
The event conventions say GenAI instrumentations “MAY capture user inputs sent to the model and responses received from it as events.” They also define gen_ai.evaluation.result for evaluation of output quality, accuracy, or other characteristics, with a recommended relationship to the evaluated operation span when possible. The events documentation says the conventions are in development and unavailable in some languages. Check current language support and compliance documentation before relying on a particular event. GenAI event conventions.
Do these 3 things before closing this tab:
1Repair Windows errors before they cause bigger problems2Scan for outdated or missing drivers - takes under a minute3Clear out junk files and repair common Windows errorsInput, output, and system-instruction capture should be an explicit application decision. The conventions permit some capture; they do not make it mandatory or establish a universal retention policy for every application.
Quick Recap
A practical implementation sequence
- Check status and support. Review the current GenAI conventions and language-specific support, since the documentation is marked Development as of October 4, 2026.
- Instrument the invocation boundary. Use a client span for a remote agent or an internal span for an in-process invocation; name the operation
invoke_agent. - Add available identity. Record agent name, stable ID, and version only when applicable and available. Keep system-instruction capture opt-in.
- Represent identifiable work. Add planning only when distinguishable; model workflow and tool operations according to their actual relationships.
- Set provider information coherently. Align
gen_ai.provider.namewith the provider-specific telemetry attributes your instrumentation actually uses. - Validate metric boundaries. Confirm invocation scope, sub-agent attribution, failed-call treatment, and exclusion of provider-side tools before comparing counts.
- Make content events deliberate. Check event support in the target language and decide explicitly whether inputs or outputs should be captured.
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
