Chronos Forecasting
Forecasting shouldn’t require an ML pipeline. The Chronos Forecasting Plugin brings zero-shot forecasting into InfluxDB 3, using Amazon’s pre-trained Chronos models to predict what’s next with no training required. Point it at a measurement, set a horizon, and get median forecasts with 50% and 80% prediction intervals, on a schedule or on demand over HTTP.
Configuration
Note: This plugin requires InfluxDB 3.8.2 or later.
Plugin parameters may be specified as key-value pairs in the --trigger-arguments flag (CLI) or in the trigger_arguments field (API) when creating a trigger.
Some plugins support TOML configuration files, which can be specified using the plugin’s config_file_path parameter.
Plugin metadata
This plugin includes a JSON metadata schema in its docstring that defines supported trigger types and configuration parameters. This metadata enables the InfluxDB 3 Explorer UI to display and configure the plugin.
Scheduled trigger parameters
| Parameter | Type | Default | Description |
|---|---|---|---|
measurement |
string | required | Source table containing historical time-series data |
field |
string | required | Numeric field name to forecast |
window |
string | required | Historical lookback window. Format: |
horizon |
int | required | Number of forecast steps to generate |
target_measurement |
string | _forecasts.{measurement} |
Destination table for forecast results |
model_id |
string | amazon/chronos-bolt-tiny |
HuggingFace model ID |
context_limit |
int | 512 |
Maximum data points fed to the model |
agg_interval |
string | 30s |
Aggregation interval for date_bin query |
tag_values |
string | none | Dot-separated tag filters, values joined by @ (e.g. tag:[email protected]:v3) |
covariate_fields |
string | none | Space-separated covariate field names. Setting this enables Chronos-2 multivariate forecasting (requires a Chronos-2 model) |
covariate_mode |
string | covariate |
How covariates are used (Chronos-2): covariate (auxiliary past covariates) or target (jointly forecast all series) |
target_database |
string | current | Database for forecast storage |
HTTP trigger parameters
HTTP parameters are sent in the JSON request body. Any value also set as a trigger argument is used as a default and overridden by the request body. The covariate_fields value may be a space-separated string or a JSON array.
| Parameter | Type | Default | Description |
|---|---|---|---|
table |
string | required | Source table name containing historical data |
field |
string | required | Numeric field name to forecast |
horizon |
int | 64 |
Number of forecast steps to generate |
context_limit |
int | 512 |
Maximum context window size (data points) |
model_id |
string | amazon/chronos-bolt-tiny |
HuggingFace model ID |
covariate_fields |
string | none | Space-separated covariate field names. Setting this enables Chronos-2 multivariate forecasting (requires a Chronos-2 model) |
covariate_mode |
string | covariate |
How covariates are used (Chronos-2): covariate (auxiliary past covariates) or target (jointly forecast all series) |
write_results |
string | false |
Write forecast results to the database |
target_measurement |
string | none | Destination table for results (required if write_results is true) |
target_database |
string | current | Database for forecast storage |
where_clause: An optional SQLWHEREclause for filtering source data, passed in the request body like any other parameter. Example:{"where_clause": "host = 'server1'"}.
TOML configuration
| Parameter | Type | Default | Description |
|---|---|---|---|
config_file_path |
string | none | Path to a TOML config file: absolute, or relative to the plugin directory (INFLUXDB3_PLUGIN_DIR or PLUGIN_DIR). Required for TOML configuration |
To use a TOML configuration file, specify the config_file_path in the trigger arguments. Relative paths are resolved from the plugin directory (INFLUXDB3_PLUGIN_DIR or PLUGIN_DIR), with a fallback to the processing engine’s virtual environment; absolute paths are used as-is.
Example TOML configuration
Ready to get started?
Download InfluxDB 3 and have running in minutes.
