CLI Reference

This file is generated from the canonical repository documentation. Do not edit it directly.

Generated from internal/cli. The serve-ops command is an accepted alias for serve; new usage should use serve.

`akritas serve`

Flag	Default	Description
`-address`	`127.0.0.1:8090`	HTTP listen address
`-alert-sources-config`	`(empty)`	provider alert source configuration; empty disables provider-specific webhook routes
`-alert-store`	`data/audit/alerts.jsonl`	append-only JSONL alert, incident and investigation-job store; empty disables alert ingestion
`-api-key-env`	`AKRITAS_API_KEY`	environment variable containing inbound Bearer API key
`-audit-log`	`data/audit/akritas.jsonl`	append-only JSONL run and audit log; empty disables persistence
`-base-url`	`http://127.0.0.1:8080/v1`	upstream OpenAI-compatible API ending in /v1
`-change-validator`	`(repeatable)`	opt-in executable validator: go-vet, go-test or yamllint; may be repeated
`-config`	`(empty)`	strict JSON service configuration; AKRITAS_CONFIG provides the default path
`-max-context-tokens`	`40000`	conservative host-counted context token limit per Run
`-max-iterations`	`12`	maximum upstream model calls per Run
`-max-model-tokens`	`16384`	maximum aggregate model output tokens per Run
`-max-retrieved-context-bytes`	`131072`	maximum retrieved context bytes per Run
`-max-tokens`	`1024`	default response token limit
`-max-tokens-limit`	`4096`	hard response token limit
`-max-tool-calls`	`8`	maximum read-only tool calls per chat request
`-max-tool-result-bytes`	`2097152`	maximum aggregate tool result bytes per Run
`-mcp-config`	`(empty)`	MCP configuration; only authorized read tools are exposed
`-model`	`akritas`	model ID exposed by this server
`-notifications-config`	`(empty)`	incident notification and bot adapter configuration; empty disables both
`-public-url`	`(empty)`	externally reachable Akritas base URL used in notification links
`-rag-index`	`(empty)`	local RAG index; empty disables RAG
`-request-timeout`	`10m`	upstream request and generation timeout
`-response-language`	`en`	BCP 47 language tag for model-generated prose; the Web UI remains English
`-result-runes`	`800`	maximum runes in each RAG excerpt
`-search-top-k`	`5`	default and maximum RAG results
`-skills-dir`	`skills`	directory of selectively loaded */SKILL.md operational skills; empty disables skills
`-system-instructions`	`instructions/SYSTEM.md`	UTF-8 Markdown file with global model instructions
`-temperature`	`0.2`	default model sampling temperature
`-upstream-api-key-env`	`OPENAI_API_KEY`	environment variable containing upstream API key
`-upstream-model`	`(empty)`	upstream model ID; empty discovers the first model
`-workspace`	`(repeatable)`	allowed change workspace as name=directory; flag may be repeated
`-workspace-config`	`(empty)`	strict JSON workspace catalog; roots are relative to the config file

`akritas serve-ops`

Flag	Default	Description
`-address`	`127.0.0.1:8090`	HTTP listen address
`-alert-sources-config`	`(empty)`	provider alert source configuration; empty disables provider-specific webhook routes
`-alert-store`	`data/audit/alerts.jsonl`	append-only JSONL alert, incident and investigation-job store; empty disables alert ingestion
`-api-key-env`	`AKRITAS_API_KEY`	environment variable containing inbound Bearer API key
`-audit-log`	`data/audit/akritas.jsonl`	append-only JSONL run and audit log; empty disables persistence
`-base-url`	`http://127.0.0.1:8080/v1`	upstream OpenAI-compatible API ending in /v1
`-change-validator`	`(repeatable)`	opt-in executable validator: go-vet, go-test or yamllint; may be repeated
`-config`	`(empty)`	strict JSON service configuration; AKRITAS_CONFIG provides the default path
`-max-context-tokens`	`40000`	conservative host-counted context token limit per Run
`-max-iterations`	`12`	maximum upstream model calls per Run
`-max-model-tokens`	`16384`	maximum aggregate model output tokens per Run
`-max-retrieved-context-bytes`	`131072`	maximum retrieved context bytes per Run
`-max-tokens`	`1024`	default response token limit
`-max-tokens-limit`	`4096`	hard response token limit
`-max-tool-calls`	`8`	maximum read-only tool calls per chat request
`-max-tool-result-bytes`	`2097152`	maximum aggregate tool result bytes per Run
`-mcp-config`	`(empty)`	MCP configuration; only authorized read tools are exposed
`-model`	`akritas`	model ID exposed by this server
`-notifications-config`	`(empty)`	incident notification and bot adapter configuration; empty disables both
`-public-url`	`(empty)`	externally reachable Akritas base URL used in notification links
`-rag-index`	`(empty)`	local RAG index; empty disables RAG
`-request-timeout`	`10m`	upstream request and generation timeout
`-response-language`	`en`	BCP 47 language tag for model-generated prose; the Web UI remains English
`-result-runes`	`800`	maximum runes in each RAG excerpt
`-search-top-k`	`5`	default and maximum RAG results
`-skills-dir`	`skills`	directory of selectively loaded */SKILL.md operational skills; empty disables skills
`-system-instructions`	`instructions/SYSTEM.md`	UTF-8 Markdown file with global model instructions
`-temperature`	`0.2`	default model sampling temperature
`-upstream-api-key-env`	`OPENAI_API_KEY`	environment variable containing upstream API key
`-upstream-model`	`(empty)`	upstream model ID; empty discovers the first model
`-workspace`	`(repeatable)`	allowed change workspace as name=directory; flag may be repeated
`-workspace-config`	`(empty)`	strict JSON workspace catalog; roots are relative to the config file

`akritas simulate-change`

Flag	Default	Description
`-api-key-env`	`OPENAI_API_KEY`	environment variable containing the API key; empty disables authentication
`-base-url`	`http://127.0.0.1:8080/v1`	OpenAI-compatible API base URL ending in /v1
`-file`	`(repeatable)`	repository file relative to root; flag may be repeated
`-max-tokens`	`1600`	maximum response tokens
`-model`	`(empty)`	model ID; empty discovers the first /v1/models entry
`-print-prompt`	`false`	print the exact system and user prompts without calling the model
`-request`	`(empty)`	change request file relative to root
`-request-timeout`	`10m`	OpenAI HTTP request timeout
`-response-language`	`en`	BCP 47 language tag for model-generated prose
`-root`	`(empty)`	snapshot root directory
`-system-instructions`	`instructions/SYSTEM.md`	UTF-8 Markdown file with global model instructions
`-temperature`	`0`	model sampling temperature

`akritas import-local`

Flag	Default	Description
`-extensions`	`built-in text/code extensions`	comma-separated extensions, or * for every UTF-8 file
`-gzip`	`true`	compress shards with gzip
`-input`	`(empty)`	root directory recursively scanned for text files
`-license`	`(empty)`	license or ownership declaration stored in every document
`-max-documents`	`0`	maximum accepted documents; 0 means unlimited
`-maximum-file-mb`	`16`	skip individual files larger than this
`-minimum-runes`	`100`	skip normalized documents shorter than this
`-output`	`data/corpus-local`	empty output directory
`-shard-mb`	`64`	maximum uncompressed JSONL MiB per shard
`-source`	`(empty)`	stable source name stored in every document
`-validation-percent`	`2`	deterministic held-out percentage in [0,100)

`akritas build-corpus`

Flag	Default	Description
`-gzip`	`true`	compress shards with gzip
`-jsonl`	`(repeatable)`	document JSONL or JSONL.GZ input; may be repeated
`-output`	`data/corpus`	empty output directory
`-shard-mb`	`64`	maximum uncompressed JSONL MiB per shard
`-text`	`(repeatable)`	UTF-8 text file stored as one document; may be repeated

`akritas build-rag-index`

Flag	Default	Description
`-chunk-runes`	`1200`	maximum runes per searchable chunk
`-manifest`	`(empty)`	source corpus manifest
`-max-documents`	`0`	optional document limit; zero means all
`-output`	`(empty)`	binary RAG index output
`-overlap-runes`	`150`	overlap between adjacent chunks
`-verify`	`true`	verify corpus shard hashes

`akritas search-rag`

Flag	Default	Description
`-json`	`false`	print machine-readable JSON
`-load`	`(empty)`	binary RAG index
`-query`	`(empty)`	search query
`-top-k`	`5`	maximum search results

`akritas inspect-rag-index`

Flag	Default	Description
`-json`	`false`	print machine-readable JSON
`-load`	`(empty)`	binary RAG index

`akritas list-rag-documents`

Flag	Default	Description
`-id`	`(empty)`	case-insensitive document ID substring
`-json`	`false`	print machine-readable JSON
`-limit`	`50`	maximum documents to print, up to 10000
`-load`	`(empty)`	binary RAG index
`-offset`	`0`	number of matching documents to skip
`-preview-runes`	`0`	optional first-chunk preview length
`-source`	`(empty)`	case-insensitive source substring
`-title`	`(empty)`	case-insensitive title substring

`akritas mcp-victoriametrics`

Flag	Default	Description
`-account-id`	`(empty)`	optional VictoriaMetrics AccountID header
`-base-url`	`http://127.0.0.1:8428/prometheus`	VictoriaMetrics read API root before /api/v1
`-bearer-token-env`	`(empty)`	environment variable containing a bearer token
`-debug`	`false`	log bounded VictoriaMetrics error response details to stderr
`-max-response-bytes`	`192*1024`	maximum decompressed VictoriaMetrics JSON response size
`-password-env`	`(empty)`	environment variable containing a basic-auth password
`-project-id`	`(empty)`	optional VictoriaMetrics ProjectID header
`-request-timeout`	`15s`	timeout for each VictoriaMetrics HTTP request
`-username-env`	`(empty)`	environment variable containing a basic-auth username

akritas serve​

akritas serve-ops​

akritas simulate-change​

akritas import-local​

akritas build-corpus​

akritas build-rag-index​

akritas search-rag​

akritas inspect-rag-index​

akritas list-rag-documents​

akritas mcp-victoriametrics​

`akritas serve`

`akritas serve-ops`

`akritas simulate-change`

`akritas import-local`

`akritas build-corpus`

`akritas build-rag-index`

`akritas search-rag`

`akritas inspect-rag-index`

`akritas list-rag-documents`

`akritas mcp-victoriametrics`