For deployment onboarding, use automatic Firebase, Supabase, or Vercel setup, or start with Manual deployment setup for other providers.
A trusted Node.js application writes spans directly to object storage with
createFilesSpanExporter() and a Files SDK bucket adapter.
Node.js application -> object storage -> npx agentpond sync -> local DuckDB
No AgentPond HTTP service is required. See Direct object-store export.
Applications send OTLP or Langfuse-compatible requests to an AgentPond service, which writes to object storage.
Application -> AgentPond HTTP service -> object storage -> npx agentpond sync -> local DuckDB
Supported endpoints:
POST /api/public/ingestionPOST /api/public/otel/v1/traces
- Object store: S3 or an S3-compatible provider
- Direct adapter:
s3fromfiles-sdk/s3 - HTTP ingestion: the AgentPond Docker image
- Container targets: ECS, EKS, App Runner, or another container runtime
Initialize the CLI configuration with:
npx agentpond env init production --provider s3 --bucket agentpondSet FILES_SDK_PROVIDER=s3 and AGENTPOND_FILES_BUCKET=agentpond on the
container. AWS credentials, region, and optional compatible endpoint remain
ambient AWS environment variables. There is no AgentPond AWS Lambda handler.
- Object store: Google Cloud Storage
- Direct adapter:
gcsfromfiles-sdk/gcs - HTTP ingestion: the AgentPond Docker image
- Container targets: Cloud Run, GKE, or another container runtime
npx agentpond env init production --provider gcs --bucket agentpondSet FILES_SDK_PROVIDER=gcs and AGENTPOND_FILES_BUCKET=agentpond on the
container and use Application Default Credentials or a service account. There
is no generic AgentPond Google Cloud Function handler.
- Object store: any Files SDK provider whose adapter uses a bucket
- Direct exporter:
createFilesSpanExporter()from@agentpond/files-sdk/otel - CLI configuration: persistent provider and bucket selection
- Credentials: provider-specific process environment variables
npx agentpond env init production \
--provider r2 \
--bucket agentpondProviders that declare an endpoint or region receive matching --endpoint and
--region values. AgentPond discovers supported providers from Files SDK at
runtime, excludes Bun-only and non-bucket adapters, and ships their peer SDKs in
the CLI and ingest image.
- Object store: private Vercel Blob
- Direct exporter:
createVercelSpanExporter()from@agentpond/vercel - Runtime: trusted Node.js Vercel Functions, routes, or server actions; not Edge, middleware, client, or static code
- Credentials: Vercel OIDC where available, otherwise
BLOB_READ_WRITE_TOKEN - Isolation:
agentpond/otel/<vercel-project-id>-<target>/in a shared private store
npx agentpond init --platform vercel
npx agentpond env use staging
npx agentpond syncVercel uses direct span export and does not require or create an AgentPond ingestion route. Production is the default target.
- Object store: dedicated private Supabase Storage bucket named
agentpond - Direct exporter:
createSupabaseSpanExporter()from@agentpond/supabase - Runtime: Supabase Edge Functions or trusted Node.js backends
- Credentials: modern Supabase secret key, with temporary legacy
service_rolecompatibility - Isolation:
otel/<project-ref>/; branches use their distinct hosted project refs
npx agentpond init --platform supabase
npx agentpond env use <project-ref>
npx agentpond syncSupabase uses direct span export and does not require an AgentPond ingestion
route. Review storage.objects RLS policies even when the bucket is private.
Run ghcr.io/marcusschiesser/agentpond on any container platform and configure
FILES_SDK_PROVIDER, AGENTPOND_FILES_BUCKET, optional
FILES_SDK_ENDPOINT/FILES_SDK_REGION, provider credentials, and
Langfuse-compatible ingestion authentication directly on the deployment.
npx agentpond dev writes directly to its local DuckDB cache and has no object
store; see Local testing.