LiyaEngine

featureConfig is the feature flag system for your tenant. It controls which capabilities are active and which intents or actions are accessible.

Schema

{
  "featureConfig": {
    "enable_session_memory": true,
    "enable_embeddings": false,
    "enable_rag": false,
    "enable_streaming": false,
    "enable_employer_features": true,
    "enable_coaching": true,
    "enable_mock_interviews": true,
    "enabled_intents": null,
    "disabled_intents": [],
    "premium_intents": ["mock_interview", "coaching_session", "candidate_prescreen", "candidate_scoring"]
  }
}

Capability flags

`enable_session_memory`

When true, the engine persists conversation history across turns within a session. Required for multi-turn intents like mock-interview and coaching-session. Default: true.

`enable_embeddings`

When true, document embeddings are generated and stored. Required prerequisite for enable_rag. Default: false.

`enable_rag`

When true, retrieval-augmented generation is active — relevant knowledge base chunks are retrieved and injected into the LLM context. Requires enable_embeddings: true. Default: false.

`enable_streaming`

When true, streaming responses are available via the preferences.stream: true request option. Default: false.

`enable_employer_features`

When true, employer-facing intents are accessible: candidate_prescreen, candidate_scoring, jd_analysis, knockout_question_generation, interview_question_generation. Default: true.

`enable_coaching`

Controls access to coaching_session. Default: true.

`enable_mock_interviews`

Controls access to mock_interview. Default: true.

Intent access control

`enabled_intents`

Whitelist of intent keys. When set, only these intents are accessible. null means all intents are enabled.

{ "enabled_intents": ["resume_analysis", "job_fit_analysis"] }

`disabled_intents`

Blacklist of intent keys. Takes precedence over enabled_intents.

{ "disabled_intents": ["general_chat"] }

`premium_intents`

List of intents that use premium_model in hybrid strategy. Override the defaults here if your use case differs.

Agentic mode

Agentic execution (multi-step tool use) is a single on/off switch per tenant — enable_agentic in featureConfig:

{ "featureConfig": { "enable_agentic": true } }

Default: true. When false, agent-runtime requests are rejected (403 FEATURE_NOT_ENABLED).

There is currently no per-tenant agenticConfig block to tune step count, timeout, or fallback behaviour — those are fixed platform-wide (mode: "auto", max_steps: 3, timeout_seconds: 45, with fallback on error/timeout/max-steps all enabled) for every tenant.

Feature Configuration