Sending Data In
Observability starts with getting your AI activity into ANTS Platform. Pick the method that fits your stack. All methods authenticate with a project key pair and produce the same traces and observations.
Create a key pair under Project Settings > API Keys (see API Keys), then configure your SDK or exporter with it. See Authentication.
1. SDKs
The SDKs give you decorators or wrappers, context managers, agent identity and automatic instrumentation of OpenAI and LangChain. The Python and JS/TS SDKs send data over OpenTelemetry.
2. OpenTelemetry (OTLP)
If you already emit OpenTelemetry, point an OTLP/HTTP exporter at
https://<your-ants-host>/api/public/otel. Spans that follow the GenAI
semantic conventions are mapped to models, tokens and cost. Claude Code, GitHub
Copilot, OpenAI Codex, Claude Cowork and CrewAI have built-in parsers.
3. Framework integrations
4. Trace connectors
Import traces from agents hosted in AWS Bedrock and Google Vertex AI without changing their code. See Trace Connectors.
5. Ingestion API
The SDKs and OTLP cover most needs. For custom pipelines, send batches of
events to POST /api/public/ingestion with Basic authentication. A request
can be up to 4.5 MB and returns 207 with a success or error per event. See
the API Reference.
What to send
To get the most out of observability, include:
- Session ID: to group traces into sessions.
- User ID: for per-user analytics.
- Agent name: to tie traces to an agent in the AI Agent Center and agent-level governance views. The Python SDK requires it on every root span.
- Tags, metadata and environment: to filter and slice traces. Environment names must be lowercase.
- Model and token usage on generations: for accurate cost. Costs are computed from your model definitions when you do not send them.
Processing and limits
- Data is processed asynchronously. New traces usually appear within seconds.
- Ingestion uses credits on ANTS-hosted plans.
When ingestion is blocked for billing reasons, the API returns
402. - ANTS-hosted deployments apply per-organization rate limits. Over the limit,
the API returns
429. - Stored data follows your organization's data retention period.
Next steps
- Traces & Observations: explore what you sent.
- Integrations: the full list of SDKs and connectors.
- API Reference: the ingestion endpoint and REST API.