Lightspeed Core Stack

📋 Configuration schema

A2AStateConfiguration

A2A protocol persistent state configuration.

Configures how A2A task state and context-to-conversation mappings are stored. For multi-worker deployments, use SQLite or PostgreSQL to ensure state is shared across all workers.

If no configuration is provided, in-memory storage is used (default). This is suitable for single-worker deployments but state will be lost on restarts and not shared across workers.

Attributes: sqlite: SQLite database configuration for A2A state storage. postgres: PostgreSQL database configuration for A2A state storage.

APIKeyTokenConfiguration

API Key Token configuration.

AccessRule

Rule defining what actions a role can perform.

Action

Available actions in the system.

Note: this is not a real model, just an enumeration of all action names.

ApprovalFilter

Granular approval control for specific MCP tools.

Attributes: always: Tool names that always require human approval before execution. never: Tool names that never require approval (pre-approved).

ApprovalsConfiguration

Configuration for human-in-the-loop approvals.

Attributes: approval_timeout_seconds: How long approval requests remain pending before expiring. approval_retention_days: How long to retain decided approvals for audit purposes before cleanup.

AuthenticationConfiguration

Authentication configuration.

AuthorizationConfiguration

Authorization configuration.

AzureEntraIdConfiguration

Microsoft Entra ID authentication attributes for Azure.

ByokRag

BYOK (Bring Your Own Knowledge) RAG configuration.

CORSConfiguration

CORS configuration.

CORS or ‘Cross-Origin Resource Sharing’ refers to the situations when a frontend running in a browser has JavaScript code that communicates with a backend, and the backend is in a different ‘origin’ than the frontend.

Useful resources:

• CORS in FastAPI
• Wikipedia article
• What is CORS?

CompactionConfiguration

Configuration for conversation history compaction.

Compaction summarizes older conversation turns when their estimated token count approaches the context window limit, keeping the conversation usable instead of failing with HTTP 413. The configuration here controls when compaction triggers and how much recent context is preserved verbatim.

Attributes: enabled: Master switch. When False, compaction never triggers and other fields are inert. threshold_ratio: Trigger compaction when estimated input tokens exceed this fraction of the model’s context window (clamped to 0.0..1.0). token_floor: Minimum estimated token count before compaction can trigger, regardless of threshold_ratio. Prevents triggering on very small context windows. buffer_turns: Initial number of recent turns to keep verbatim. The runtime applies a degrading guard — if these turns exceed the available budget, it reduces buffer_turns by one repeatedly until the budget fits, down to zero. buffer_max_ratio: Hard cap on the fraction of the context window the buffer zone may occupy, regardless of buffer_turns.

Configuration

Global service configuration.

ConversationHistoryConfiguration

Conversation history configuration.

CustomProfile

Custom profile customization for prompts and validation.

Customization

Service customization.

DatabaseConfiguration

Database configuration.

InMemoryCacheConfig

In-memory cache configuration.

InferenceConfiguration

Inference configuration.

JsonPathOperator

Supported operators for JSONPath evaluation.

Note: this is not a real model, just an enumeration of all supported JSONPath operators.

JwkConfiguration

JWK (JSON Web Key) configuration.

A JSON Web Key (JWK) is a JavaScript Object Notation (JSON) data structure that represents a cryptographic key.

Useful resources:

• JSON Web Key
• RFC 7517

JwtConfiguration

JWT (JSON Web Token) configuration.

JSON Web Token (JWT) is a compact, URL-safe means of representing claims to be transferred between two parties. The claims in a JWT are encoded as a JSON object that is used as the payload of a JSON Web Signature (JWS) structure or as the plaintext of a JSON Web Encryption (JWE) structure, enabling the claims to be digitally signed or integrity protected with a Message Authentication Code (MAC) and/or encrypted.

Useful resources:

• JSON Web Token
• RFC 7519
• JSON Web Tokens

JwtRoleRule

Rule for extracting roles from JWT claims.

LlamaStackConfiguration

Llama stack configuration.

Llama Stack is a comprehensive system that provides a uniform set of tools for building, scaling, and deploying generative AI applications, enabling developers to create, integrate, and orchestrate multiple AI services and capabilities into an adaptable setup.

Useful resources:

• Llama Stack
• Python Llama Stack client
• Build AI Applications with Llama Stack

ModelContextProtocolServer

Model context protocol server configuration.

MCP (Model Context Protocol) servers provide tools and capabilities to the AI agents. These are configured by this structure. Only MCP servers defined in the lightspeed-stack.yaml configuration are available to the agents. Tools configured in the llama-stack run.yaml are not accessible to lightspeed-core agents.

Useful resources:

• Model Context Protocol
• MCP FAQs
• Wikipedia article

OkpConfiguration

OKP (Offline Knowledge Portal) provider configuration.

Controls provider-specific behaviour for the OKP vector store. Only relevant when "okp" is listed in rag.inline or rag.tool.

PostgreSQLDatabaseConfiguration

PostgreSQL database configuration.

PostgreSQL database is used by Lightspeed Core Stack service for storing information about conversation IDs. It can also be leveraged to store conversation history and information about quota usage.

Useful resources:

• Psycopg: connection classes
• PostgreSQL connection strings
• How to Use PostgreSQL in Python

QuotaHandlersConfiguration

Quota limiter configuration.

It is possible to limit quota usage per user or per service or services (that typically run in one cluster). Each limit is configured as a separate quota limiter. It can be of type user_limiter or cluster_limiter (which is name that makes sense in OpenShift deployment).

QuotaLimiterConfiguration

Configuration for one quota limiter.

There are three configuration options for each limiter:

1. period is specified in a human-readable form, see https://www.postgresql.org/docs/current/datatype-datetime.html#DATATYPE-INTERVAL-INPUT for all possible options. When the end of the period is reached, the quota is reset or increased.
2. initial_quota is the value set at the beginning of the period.
3. quota_increase is the value (if specified) used to increase the quota when the period is reached.

There are two basic use cases:

1. When the quota needs to be reset to a specific value periodically (for example on a weekly or monthly basis), set initial_quota to the required value.
2. When the quota needs to be increased by a specific value periodically (for example on a daily basis), set quota_increase.

QuotaSchedulerConfiguration

Quota scheduler configuration.

RHIdentityConfiguration

Red Hat Identity authentication configuration.

RagConfiguration

RAG strategy configuration.

Controls which RAG sources are used for inline and tool-based retrieval.

Each strategy lists RAG IDs to include. The special ID "okp" defined in constants, activates the OKP provider; all other IDs refer to entries in byok_rag.

Backward compatibility: - inline defaults to [] (no inline RAG). - tool defaults to [] (no tool RAG).

If no RAG strategy is defined (inline and tool are empty), the RAG tool will register all stores available to llama-stack.

RerankerConfiguration

Reranker configuration for RAG chunk reranking.

RlsapiV1Configuration

Configuration for the rlsapi v1 /infer endpoint.

Settings specific to the RHEL Lightspeed Command Line Assistant (CLA) stateless inference endpoint. Kept separate from shared configuration sections so that CLA-specific options do not affect other endpoints.

SQLiteDatabaseConfiguration

SQLite database configuration.

ServiceConfiguration

Service configuration.

Lightspeed Core Stack is a REST API service that accepts requests on a specified hostname and port. It is also possible to enable authentication and specify the number of Uvicorn workers. When more workers are specified, the service can handle requests concurrently.

SkillsConfiguration

Agent skills configuration.

Specifies paths to skill directories. Skill metadata (name, description) is read from SKILL.md frontmatter at startup.

Each path can point to either: - A directory containing a SKILL.md file (single skill) - A directory containing subdirectories with SKILL.md files (multiple skills)

Paths are validated at startup to ensure they exist and contain valid SKILL.md files.

SplunkConfiguration

Splunk HEC (HTTP Event Collector) configuration.

Splunk HEC allows sending events directly to Splunk over HTTP/HTTPS. This configuration is used to send telemetry events for inference requests to the corporate Splunk deployment.

Useful resources:

• Splunk HEC Docs
• About HEC

TLSConfiguration

TLS configuration.

Transport Layer Security (TLS) is a cryptographic protocol designed to provide communications security over a computer network, such as the Internet. The protocol is widely used in applications such as email, instant messaging, and voice over IP, but its use in securing HTTPS remains the most publicly visible.

Useful resources:

• FastAPI HTTPS Deployment
• Transport Layer Security Overview
• What is TLS

TrustedProxyConfiguration

Configuration for trusted-proxy auth module.

TrustedProxyServiceAccount

A Kubernetes ServiceAccount identity for trusted-proxy allowlist.

UnifiedInferenceProvider

A high-level inference provider entry for unified-mode synthesis.

Operators describe inference providers at this high level (backend-agnostic vocabulary) instead of authoring raw Llama Stack provider blocks. The synthesizer (apply_high_level_inference) expands each entry into a Llama Stack providers.inference entry, mapping type to a provider_type and emitting ${env.<VAR>} references for secrets (never literal values).

Attributes: type: Canonical provider identifier. Vendor-neutral so it survives a future backend change; each backend-specific synthesizer maps it to its own provider vocabulary. api_key_env: Name of the environment variable holding the provider API key. Emitted verbatim as ${env.<name>} so the secret never lands on disk resolved. allowed_models: Optional allow-list of model identifiers passed through to the synthesized provider config. extra: Additional provider-config keys merged verbatim into the synthesized provider’s config block — an escape hatch for provider-specific knobs not modeled here.

UnifiedLlamaStackConfig

Backend-specific knobs for unified-mode Llama Stack synthesis.

Per Decision S5 of the design spike, backend-agnostic high-level sections (inference, …) live at the configuration root, not here. This block holds only the Llama-Stack-specific synthesis controls: which baseline to start from, an optional profile file, and a raw native_override escape hatch.

Attributes: baseline: Synthesis starting point. “default” begins from LCORE’s built-in baseline (src/data/default_run.yaml); “empty” begins from an empty dict (used by the migration tool for an exact round-trip). Ignored when profile is set. profile: Optional path to a user-authored run.yaml-shaped file used as the synthesis baseline. Relative paths resolve against the directory of the loaded lightspeed-stack.yaml. native_override: Raw Llama Stack schema deep-merged last (maps merge recursively, lists and scalars replace). The escape hatch for anything the high-level sections do not express.

UserDataCollection

User data collection configuration.