Configuration

Configuration Reference¶

Use this page as the complete reference for settings loaded by redis_sre_agent/core/config.py.

For configuration precedence, .env behavior, config-file discovery order, and setup examples, see Configuration How-to.

Config File Selector¶

Setting	Environment Variable	Type	Default	Notes
Config file path	`SRE_AGENT_CONFIG`	`str`	unset	Optional path to a YAML, TOML, or JSON config file. If unset, the app checks `config.yaml`, `config.yml`, `config.toml`, `config.json`, `sre_agent_config.yaml`, `sre_agent_config.yml`, `sre_agent_config.toml`, `sre_agent_config.json`.

Application¶

Field	Environment Variable	Type	Default	Notes
`app_name`	`APP_NAME`	`str`	`Redis SRE Agent`	Application name.
`debug`	`DEBUG`	`bool`	`false`	Debug mode toggle.
`log_level`	`LOG_LEVEL`	`str`	`INFO`	Logging level.

Server¶

Field	Environment Variable	Type	Default	Notes
`host`	`HOST`	`str`	`0.0.0.0`	API bind host.
`port`	`PORT`	`int`	`8000`	API bind port.

Redis¶

Field	Environment Variable	Type	Default	Notes
`redis_url`	`REDIS_URL`	`SecretStr`	`redis://localhost:7843/0`	Redis connection URL for agent data.

LLM / OpenAI¶

Field	Environment Variable	Type	Default	Notes
`openai_api_key`	`OPENAI_API_KEY`	`str \\| None`	`None`	OpenAI API key.
`openai_base_url`	`OPENAI_BASE_URL`	`str \\| None`	`None`	Optional OpenAI-compatible base URL.
`openai_model`	`OPENAI_MODEL`	`str`	`gpt-5`	Main reasoning model.
`openai_model_mini`	`OPENAI_MODEL_MINI`	`str`	`gpt-5-mini`	Lightweight model for utility/search tasks.
`openai_model_nano`	`OPENAI_MODEL_NANO`	`str`	`gpt-5-nano`	Smallest model for simple classification/triage.

Embeddings / Vector Search¶

Field	Environment Variable	Type	Default	Notes
`embedding_provider`	`EMBEDDING_PROVIDER`	`str`	`openai`	`openai` or `local`.
`embedding_model`	`EMBEDDING_MODEL`	`str`	`text-embedding-3-small`	Embedding model name.
`vector_dim`	`VECTOR_DIM`	`int`	`1536`	Must match embedding model output dimensions.
`embeddings_cache_ttl`	`EMBEDDINGS_CACHE_TTL`	`int \\| None`	`604800`	Embedding cache TTL in seconds; `None` means no expiration.

Task Queue¶

Field	Environment Variable	Type	Default	Notes
`task_queue_name`	`TASK_QUEUE_NAME`	`str`	`sre_agent_tasks`	Docket queue name.
`max_task_retries`	`MAX_TASK_RETRIES`	`int`	`3`	Maximum retries per task.
`task_timeout`	`TASK_TIMEOUT`	`int`	`1200`	Task timeout in seconds.

Agent Runtime¶

Field	Environment Variable	Type	Default	Notes
`max_iterations`	`MAX_ITERATIONS`	`int`	`50`	Max iterations for the main agent.
`knowledge_max_iterations`	`KNOWLEDGE_MAX_ITERATIONS`	`int`	`8`	Max iterations for the knowledge-only agent.
`max_tool_calls_per_stage`	`MAX_TOOL_CALLS_PER_STAGE`	`int`	`3`	Max tool/knowledge calls per worker stage.
`max_recommendation_topics`	`MAX_RECOMMENDATION_TOPICS`	`int`	`3`	Max recommendation topics processed per request.
`max_rejections`	`MAX_REJECTIONS`	`int`	`1`	Max correction attempts from safety/fact-check rejections.
`recursion_limit`	`RECURSION_LIMIT`	`int`	`100`	LangGraph recursion limit.
`tool_timeout`	`TOOL_TIMEOUT`	`int`	`60`	Tool execution timeout in seconds.

Tool Caching¶

Field	Environment Variable	Type	Default	Notes
`tool_cache_enabled`	`TOOL_CACHE_ENABLED`	`bool`	`true`	Enables Redis-backed tool output cache.
`tool_cache_default_ttl`	`TOOL_CACHE_DEFAULT_TTL`	`int`	`60`	Default tool cache TTL in seconds.
`tool_cache_ttl_overrides`	`TOOL_CACHE_TTL_OVERRIDES`	`dict[str, int]`	`{}`	JSON map of per-tool TTL overrides. Example: `{"info": 120}`.

LLM Retry / Timeout / Factories¶

Field	Environment Variable	Type	Default	Notes
`llm_max_retries`	`LLM_MAX_RETRIES`	`int`	`3`	Retry attempts for LLM requests.
`llm_initial_delay`	`LLM_INITIAL_DELAY`	`float`	`1.0`	Initial retry delay (seconds).
`llm_backoff_factor`	`LLM_BACKOFF_FACTOR`	`float`	`2.0`	Exponential backoff multiplier.
`llm_timeout`	`LLM_TIMEOUT`	`float`	`180.0`	LLM HTTP timeout in seconds.
`llm_factory`	`LLM_FACTORY`	`str \\| None`	`None`	Dot-path to custom LangChain chat model factory.
`async_openai_client_factory`	`ASYNC_OPENAI_CLIENT_FACTORY`	`str \\| None`	`None`	Dot-path to custom AsyncOpenAI-compatible client factory.

Monitoring¶

Field	Environment Variable	Type	Default	Notes
`prometheus_url`	`PROMETHEUS_URL`	`str \\| None`	`None`	Optional app-level Prometheus URL.
`grafana_url`	`GRAFANA_URL`	`str \\| None`	`None`	Optional app-level Grafana URL.
`grafana_api_key`	`GRAFANA_API_KEY`	`str \\| None`	`None`	Optional Grafana API key.

Security¶

Field	Environment Variable	Type	Default	Notes
`api_key`	`API_KEY`	`str \\| None`	`None`	API authentication key.
`allowed_hosts`	`ALLOWED_HOSTS`	`list[str]`	`["*"]`	CORS allow-list; pass as JSON in env var.

Support Package Storage¶

Field	Environment Variable	Type	Default	Notes
`support_package_artifacts_dir`	`SUPPORT_PACKAGE_ARTIFACTS_DIR`	`Path`	`/tmp/sre_agent/support_packages`	Local extraction/storage path.
`support_package_storage_type`	`SUPPORT_PACKAGE_STORAGE_TYPE`	`str`	`local`	Storage backend: `local` or `s3`.
`support_package_s3_bucket`	`SUPPORT_PACKAGE_S3_BUCKET`	`str \\| None`	`None`	S3 bucket name when using `s3` storage.
`support_package_s3_prefix`	`SUPPORT_PACKAGE_S3_PREFIX`	`str`	`support-packages/`	S3 key prefix.
`support_package_s3_region`	`SUPPORT_PACKAGE_S3_REGION`	`str \\| None`	`None`	Optional AWS region override.
`support_package_s3_endpoint`	`SUPPORT_PACKAGE_S3_ENDPOINT`	`str \\| None`	`None`	Optional custom S3 endpoint (for MinIO/S3-compatible storage).

Tool Providers¶

Field	Environment Variable	Type	Default	Notes
`tool_providers`	`TOOL_PROVIDERS`	`list[str]`	Built-in provider class list	JSON array of provider class paths to enable.

MCP Servers¶

Field	Environment Variable	Type	Default	Notes
`mcp_servers`	`MCP_SERVERS`	`dict[str, MCPServerConfig]`	Empty dict	JSON object of MCP server definitions. No MCP integrations are enabled unless you configure them.

MCPServerConfig object keys:

command: command to start MCP server (stdio mode)
args: command arguments
env: environment variables for the server process
url: URL for remote MCP server (HTTP/SSE)
headers: HTTP headers for remote MCP server
transport: streamable_http (recommended) or sse
tools: optional map of tool names to overrides

MCPToolConfig object keys inside tools:

capability: capability classification (for example logs, metrics, tickets, repos)
description: replacement description shown to the agent ({original} keeps upstream description text)

Environment Value Formats¶

For structured settings, environment variables must be JSON strings:

ALLOWED_HOSTS: JSON array, e.g. ["localhost", "127.0.0.1"]
TOOL_PROVIDERS: JSON array of class paths
TOOL_CACHE_TTL_OVERRIDES: JSON object of {tool_name: ttl_seconds}
MCP_SERVERS: JSON object matching MCPServerConfig