Configuration

Every Prologue tool binds the same cratis-prologue.json, via the shared Cratis.Prologue.Configuration package — never appsettings.json. The Extractor binds the prologue section; the Interpreter binds llm; the HTTP capture source binds the top-level reverseProxy section (a plain YARP configuration block, not a Prologue-specific schema).

Where the file comes from

• Each tool looks for cratis-prologue.json in its working directory by default. Override the path with the PROLOGUE_CONFIG environment variable, or point --config-style hosting at a different basePath if you’re calling AddPrologueConfiguration() yourself.
• The file is the baseline; environment variables override it. Every property below has a double-underscore environment variable equivalent — Prologue__Output__Kind, Prologue__SqlServer__0__ConnectionString, ReverseProxy__Clusters__monitored__Destinations__primary__Address — so a deployed tool can be configured by its host (a container, an orchestrator, an Aspire composition) without rewriting the file. If you’re hosting a Prologue tool yourself, AddPrologueConfiguration() from Cratis.Prologue.Configuration wires up that precedence for you.
• The file is missing? That’s fine — every property below has a default, and environment variables still apply.

Root shape

Property	Type	Purpose
prologue	PrologueOptions	Capture sources, correlation, and output — bound by the Extractor
llm	LlmOptions	Optional language-model refinement — bound by the Interpreter
reverseProxy	YARP configuration	Routes and clusters for the HTTP capture source — see YARP’s configuration docs

prologue

Property	Type	Default	Purpose
prologueId	guid	a fresh id	Identifies this Prologue’s captures, so a folder or Receiver can hold more than one Prologue’s data without mixing them
output	OutputOptions	—	Where captures are written
correlation	CorrelationOptions	—	The correlation time window
sqlServer	array of SqlServerOptions	[]	SQL Server databases to watch via CDC
postgres	array of PostgresOptions	[]	PostgreSQL databases to watch via logical replication
openTelemetry	OpenTelemetryOptions	—	OTLP proxy configuration

prologue.output

Property	Type	Default	Purpose
kind	"Api" | "Json"	"Api"	Whether captures are posted to a Receiver or written to JSON files
api.endpoint	string	"http://localhost:5005"	Receiver base address, used when kind is "Api"
json.directory	string	"./captures"	Folder to write rolling .jsonl capture files to, used when kind is "Json"
json.maxEntriesPerFile	int	10000	Entries per file before the Extractor rolls to a new one

prologue.correlation

Property	Type	Default	Purpose
windowMilliseconds	int	2000	How long after a command the correlator keeps grouping database transactions and telemetry into the same capture — see How Prologue works

prologue.sqlServer[]

Property	Type	Default	Purpose
name	string	"sqlserver"	Label for this database in captures
connectionString	string	—	Connection string; needs permission to enable CDC
enableChangeDataCapture	bool	true	Whether the Extractor enables CDC on the database itself
tables	array of string	[]	Table allowlist; empty captures every user table
pollIntervalMilliseconds	int	500	How often the Extractor polls for new CDC changes

prologue.postgres[]

Property	Type	Default	Purpose
name	string	"postgres"	Label for this database in captures
connectionString	string	—	Connection string; needs permission to create a replication slot and publication, and wal_level = logical on the server
slot	string	"prologue_slot"	Replication slot name
publication	string	"prologue_publication"	Publication name

prologue.openTelemetry

Property	Type	Default	Purpose
enabled	bool	false	Whether the OTLP proxy is active
serviceNames	array of string	[]	Service name allowlist; empty captures telemetry from every service
attributeKeys	array of string	[]	Span/log attribute keys whose values are captured; everything else is dropped
upstream.http	string	—	Upstream OTLP/HTTP collector to forward telemetry to
upstream.grpc	string	—	Upstream OTLP/gRPC collector to forward telemetry to

Leave both upstream addresses empty to make the Extractor a terminal collector rather than a forwarding proxy.

llm

Property	Type	Default	Purpose
enabled	bool	false	Whether the Interpreter refines the heuristic model with a language model
kind	"Ollama" | "OpenAI" | "AzureOpenAI" | "OpenAICompatible" | "Anthropic"	"Ollama"	Which provider to call
endpoint	string	"http://llm:11434"	Provider endpoint; the hosted providers (OpenAI, Azure OpenAI, Anthropic) default to their public endpoint if you don’t set one
accessToken	string	—	API key for the hosted providers; not needed for a local Ollama endpoint
modelId	string	an Ollama model tag	The model to call — for AzureOpenAI, this is the deployment name, not a model name
refinementTimeout	duration	2 minutes	How long the Interpreter waits for a refinement response before giving up and keeping the heuristic result

See Running the Interpreter for a worked example.

Next

• Capture sources — what triggers a capture per source.
• Point Prologue at your system — a complete worked cratis-prologue.json.