HTTP ingest API

Use the HTTP API when no SDK or OpenTelemetry exporter fits your stack. It has two endpoints: one for events that your code sends, and one for log lines from a host or a log forwarder.

Endpoint	Use it for
`POST https://ingest.squasher.ai/v1/ingest/<project-id>`	Errors and logs that your code sends
`POST https://ingest.squasher.ai/v1/drain/<source>/<project-id>`	Log lines from a host or a forwarder

Authenticate requests

Both endpoints read the ingest key from one of these places:

The x-squasher-key: <key> header.
The Authorization: Bearer <key> header.
The key query parameter (?key=<key>), for tools that cannot set headers.

The project ID in the path must belong to the key. If it does not, Squasher returns 403. To get both values, see Create an ingest key.

Send an event

curl -X POST "https://ingest.squasher.ai/v1/ingest/$SQUASHER_PROJECT_ID" \
  -H "Content-Type: application/json" \
  -H "x-squasher-key: $SQUASHER_API_KEY" \
  -d '{"message":"Squasher test error","type":"TestError","level":"error","environment":"production"}'

A 200 response means that Squasher accepted the event. An event at error or fatal level becomes an issue. To send more than one event, use one of these bodies:

A JSON array of event objects.
An object with an events array.
One event object on each line, with Content-Type: application/x-ndjson.

Every event needs a string message. If an event or line is not valid, Squasher rejects the full request with 400 and names the event or line.

Event fields

Field	Type	What Squasher does with it
`message`	string, required	Stores it as the log body. Issue titles come from it.
`level`	string	`fatal`, `critical`, `error`, `warning`, `warn`, `info`, or `debug`. The default is `error`.
`type`	string	Stores it as the error type. Squasher uses it to group issues.
`stack`	string	Stores the raw stack trace.
`frames`	array	Stores the parsed stack frames.
`environment`	string	Sets the environment, for example `production`.
`release`	string	Sets the release. Squasher ignores empty or invalid values.
`timestamp`	number or string	Unix seconds, milliseconds, or nanoseconds, or ISO 8601. The default is the time received.
`tags`	object	Stores string values as attributes. `service` sets the service, and `host` sets the host name.
`attributes`	object	Stores the values as attributes and keeps numbers and booleans.
`extra`	object	Stores the values as attributes.
`user`	object	Stores the object as JSON.
`request`	object	Stores the object as JSON.
`trace`	object	`trace_id` and `span_id` link the event to a trace. With `span_name`, Squasher also stores a span.
`measurements`	array	Stores each `{ "name", "value", "unit" }` item as a metric point.
`breadcrumbs`	array	Accepted, but not stored.

Automatic AI triage runs only on issues in the production environment by default. Set environment on events that must be triaged.

Send log lines to the drain endpoint

The drain endpoint takes JSON log records in any shape and keeps a few known fields. The platform guides, such as Vercel and Fly.io, use it.

curl -X POST "https://ingest.squasher.ai/v1/drain/my-forwarder/$SQUASHER_PROJECT_ID" \
  -H "x-squasher-key: $SQUASHER_API_KEY" \
  --data-binary $'{"message":"payment failed","level":"error","service":"billing","timestamp":1758800000000}\n{"message":"payment retried","level":"info","service":"billing"}'

<source> is a name that you choose, with lowercase letters, digits, _, and -. Squasher stores it on each record and uses it as the service name when a record has none.
Send a JSON array of objects, or one JSON object on each line. Squasher skips lines that are not JSON objects. A JSON array must contain only objects.
Syslog bodies also work: Heroku application/logplex-1 octet-counted frames, or one RFC 5424 message on each line. See Heroku.
The body can be gzip or deflate compressed. Squasher also detects gzip without a Content-Encoding header. Other encodings return 415. A body can be at most 16 MiB, before and after decompression. A larger body returns 413.
Add ?environment=<name> to the URL to set the environment of records that have none. The name can contain letters, digits, ., _, and -, up to 64 characters.
Squasher skips a record that has "squasher.skip_ingest": true.

What Squasher keeps from each record

Squasher field	Read from	If missing
Message	`message`, then `msg`	The full record as JSON
Level	`level`, then `severity`. See How levels become issues.	Read from the message
Time	`timestamp`: Unix seconds, milliseconds, or nanoseconds, or ISO 8601	The time received
Service	`service.name`, `serviceName`, `service_name`, `service`, `projectName`, `project`, or `fly.app.name`, also inside a JSON `message`	The `<source>` path segment
Environment	`environment`, `deployment.environment`, `deployment.environment.name`, `environmentName`, or Vercel `target`	The `?environment=` value
Deployment ID	`deploymentId`	Empty
Error type	`errorType`	Empty
Trace and span ID	`traceId` and `spanId`	Empty

Squasher drops all other fields, including nested objects such as proxy or metadata. It reads only top-level keys, except the Fly.io log.level and fly.app.name fields. Squasher reads a numeric timestamp below 10¹⁰ as seconds, below 10¹⁶ as milliseconds, and above that as nanoseconds. A timestamp in microseconds is read as milliseconds, so convert it before you send it.

How levels become issues

Level in the record	Stored as	Creates an issue
`fatal`, `critical`	Fatal	Yes
`error`	Error	Yes
`warn`, `warning`	Warning	No
`debug`	Debug	No
Any other value	Info	No

A top-level level or severity is used as sent. A record without one gets its level from, in order:

A platform level that is not info: the Fly.io log.level, or the syslog severity.
The message: a JSON "level" field (pino numbers too), level=error or at=error, the first uppercase level word such as ERROR, [WARN], or FATAL:, or a stack trace such as Traceback (most recent call last):, TypeError: ..., or panic:.
info, when nothing matches.

Squasher does not set a level from an HTTP status code, and it does not link drain records by request ID. Automatic AI triage runs only on issues with an enabled environment, so set ?environment= when your records have no environment field. Otherwise, select Run triage in the Squasher triage card on the issue to start it by hand. Squasher groups error records into issues by service, error type, and the first line of the message. Before it compares messages, it replaces numbers, hexadecimal strings, and UUIDs.

Limits

Each project can send 10,000 records per minute by default. Squasher counts each log record, span, and metric point. Above the limit, it returns 429 with Retry-After and X-RateLimit-Limit, X-RateLimit-Remaining, and X-RateLimit-Reset headers.
A 503 response with Retry-After: 1 means that ingest is not available for a moment. Retry with backoff.
The events endpoint reads the body as uncompressed text. The drain endpoint also reads gzip and deflate bodies.

For LLM and agent telemetry, use the Agent telemetry SDK. To send Langfuse-compatible batches without it, post them to POST /v1/ai/ingest/<project-id>.

Next steps

OpenTelemetry

Send traces, logs, and metrics from any language over OTLP.

Vector

Forward logs from files and containers through a Vector sink.

Node.js SDK

Capture errors with stack traces from Node.js services.

Issues

See how Squasher groups errors into issues.

Get started

Guides

Authenticate requests

Send an event

Event fields

Send log lines to the drain endpoint

What Squasher keeps from each record

How levels become issues

Limits

Next steps

OpenTelemetry

Vector

Node.js SDK

Issues

​Authenticate requests

​Send an event

​Event fields

​Send log lines to the drain endpoint

​What Squasher keeps from each record

​How levels become issues

​Limits

​Next steps

OpenTelemetry

Vector

Node.js SDK

Issues

Authenticate requests

Send an event

Event fields

Send log lines to the drain endpoint

What Squasher keeps from each record

How levels become issues

Limits

Next steps