Table Config

This page is the control-plane map for ConvEngine. It explains the important ce_* tables, what each one controls, which step reads it, and which config families shape the runtime without requiring custom Java branching.

How to use this page

If you are debugging behavior, start with the table that owns the decision. If you are tuning behavior, start with the config family that shapes the step before you touch custom code.

Quick jump

• Detailed static table reference
• Core static control tables
• What these tables actually do
• Runtime persistence tables
• Semantic Query metadata tables
• Config families and values
• High-impact lookup map

Detailed static table reference

The sections below go one table at a time for the primary static control-plane tables. Each table shows the important columns, the typical values, and what those values actually do at runtime.

ce_intent

ce_intent is the root business-intent registry. If an intent does not exist here, later scoped rows should not depend on it.

ce_intent columns

Column	Typical values	What it does
intent_code	GREETING, LOAN_APPLICATION, ORDER_DIAGNOSTICS, SEMANTIC_QUERY_DIAGNOSTICS	Primary intent identifier used everywhere else in scoped config.
description	Free text	Human-readable explanation of the intent.
priority	Integer	Used when multiple intents need deterministic ordering.
enabled	true / false	Disables the intent without deleting the row.
display_name	User-friendly text	UI/admin-facing label for the intent.
llm_hint	Prompt-style hint text	Optional LLM biasing hint used by current classifier/runtime helpers.
created_at	Timestamp	Row creation audit timestamp.

ce_intent_classifier

ce_intent_classifier is the first business-routing layer. It maps raw user text into intent and initial state.

ce_intent_classifier columns

Column	Typical values	What it does
classifier_id	Integer / PK	Unique classifier row id.
intent_code	Any enabled ce_intent.intent_code	Intent selected when the classifier row matches.
state_code	ANY, UNKNOWN, or concrete state values	Initial state set along with the intent.
rule_type	Consumer-defined pattern mode (commonly REGEX / EXACT / CONTAINS style semantics)	Controls how pattern matching should be interpreted.
pattern	Regex or literal text	The text expression checked against user input.
priority	Integer	Lower/higher priority ordering depending on consumer precedence rules.
enabled	true / false	Turns the classifier row on or off.
description	Free text	Explains why the classifier row exists.

ce_config

ce_config is the generic key/value fallback control table for framework defaults and legacy configuration paths.

ce_config columns

Column	Typical values	What it does
config_id	Integer / PK	Unique config row id.
config_type	McpPlanner, Flow, Audit, etc.	Groups related config keys.
config_key	String key	The specific setting name inside the type group.
config_value	String / JSON text	The configured value read by the framework.
enabled	true / false	Disables the config row without deleting it.
created_at	Timestamp	Row creation audit timestamp.

ce_container_config

ce_container_config maps an input parameter to a concrete page / section / container location for a given intent and state.

ce_container_config columns

Column	Typical values	What it does
id	Integer / PK	Unique row id.
intent_code	Concrete intent code	Scopes the container row to an intent.
state_code	Concrete state code	Further scopes the row to a state.
page_id	Integer page identifier	UI page identifier for the mapping.
section_id	Integer section identifier	UI section identifier within the page.
container_id	Integer container identifier	Specific container slot for the mapping.
input_param_name	customerId, accountId, accLocId	The runtime input param name associated with this UI container.
priority	Integer	Controls which rows apply first if multiple are eligible.
enabled	true / false	Disables the row without deleting it.
created_at	Timestamp	Row creation audit timestamp.

ce_output_schema

ce_output_schema defines the structured extraction and collection contract for a given intent/state.

ce_output_schema columns

Column	Typical values	What it does
schema_id	Integer / PK	Unique schema row id.
intent_code	Concrete intent code	Scopes the schema to an intent.
state_code	Concrete state	Scopes the schema to a state.
json_schema	JSON Schema text	Defines required fields, property types, and extraction contract.
description	Free text	Explains the purpose of the schema.
enabled	true / false	Turns the schema on or off.
priority	Integer	Controls schema selection order.

ce_prompt_template

ce_prompt_template now controls both prompt text and turn semantics.

ce_prompt_template columns

Column	Typical values	What it does
template_id	Integer / PK	Unique prompt row id.
intent_code	Concrete intent code	Scopes the prompt to an intent.
state_code	Concrete state	Scopes the prompt to a state.
response_type	EXACT, DERIVED and related consumer response strategies	Selects which prompt path this row belongs to.
system_prompt	Prompt text	System-level prompt text.
user_prompt	Prompt text	User-facing prompt template text.
temperature	Decimal	Optional LLM temperature override.
interaction_mode	NORMAL, IDLE, COLLECT, CONFIRM, PROCESSING, FINAL, ERROR, DISAMBIGUATE, FOLLOW_UP, PENDING_ACTION, REVIEW	Broad turn-behavior contract used by current routing.
interaction_contract	JSON text such as {"allows":[...],"expects":[...]}	Fine-grained capabilities and input expectations used by routing and correction.
enabled	true / false	Turns the prompt row on or off.
created_at	Timestamp	Row creation audit timestamp.

ce_response

ce_response is still the final response authority even when MCP or semantic query runtime computes internal answers.

ce_response columns

Column	Typical values	What it does
response_id	Integer / PK	Unique response row id.
intent_code	Concrete intent code	Scopes the response row to an intent.
state_code	Concrete state	Scopes the response row to a state.
output_format	TEXT, JSON	Controls the transport/output shape.
response_type	EXACT, DERIVED	Controls whether text is direct or derived from runtime context.
exact_text	Prompt/result text	Returned directly for EXACT responses.
derivation_hint	Path or expression hint	Tells the resolver where to derive output for DERIVED responses.
json_schema	Optional JSON schema	Output schema contract for structured responses.
priority	Integer	Controls which response row wins when multiple are eligible.
enabled	true / false	Turns the response row on or off.
description	Free text	Human-readable explanation.
created_at	Timestamp	Row creation audit timestamp.

ce_rule

ce_rule is the biggest behavior table in the framework. Most domain flow changes should start here.

ce_rule columns

Column	Typical values	What it does
rule_id	Integer / PK	Unique rule row id.
phase	POST_DIALOGUE_ACT, POST_SCHEMA_EXTRACTION, PRE_AGENT_MCP, PRE_RESPONSE_RESOLUTION, POST_AGENT_INTENT, POST_AGENT_MCP, POST_TOOL_EXECUTION	Controls when the rule is evaluated.
intent_code	Concrete intent, ANY, UNKNOWN	Scopes the rule by intent.
state_code	Concrete state, ANY, UNKNOWN	Scopes the rule by state.
rule_type	EXACT, REGEX, JSON_PATH or consumer-supported matcher mode	Controls how match_pattern is interpreted.
match_pattern	Pattern or JSON_PATH expression	The actual condition expression.
action	SET_INTENT, SET_STATE, SET_DIALOGUE_ACT, SET_JSON, GET_CONTEXT, GET_SCHEMA_JSON, GET_SESSION, SET_TASK, SET_INPUT_PARAM	The mutation or retrieval action performed on match.
action_value	State code, JSON path, input-param payload, etc.	Payload used by the selected action.
priority	Integer	Controls ordering when multiple rules match.
enabled	true / false	Turns the rule row on or off.
description	Free text	Explains rule purpose.
created_at	Timestamp	Row creation audit timestamp.

ce_policy

ce_policy is the pre-intent governance layer.

ce_policy columns

Column	Typical values	What it does
policy_id	Integer / PK	Unique policy row id.
rule_type	REGEX, EXACT, CONTAINS or consumer-defined matcher mode	Controls how the pattern is interpreted.
pattern	Literal text or regex	The condition used by the policy check.
response_text	User-facing policy response text	The response text returned when the policy row triggers.
priority	Integer	Controls ordering across policy rows.
enabled	true / false	Turns the policy row on or off.
description	Free text	Explains the policy.
created_at	Timestamp	Row creation audit timestamp.

ce_pending_action

ce_pending_action is the static task catalog. The runtime lifecycle is stored in context, not in this table.

ce_pending_action columns

Column	Typical values	What it does
pending_action_id	Integer / PK	Unique pending-action row id.
intent_code	Concrete intent code	Scopes the action to an intent.
state_code	Concrete state code	Scopes the action to a state.
action_key	Business action key	Stable identifier used during routing and execution.
bean_name	Spring bean name	The consumer bean invoked by the task executor.
method_names	Comma-separated method names or configured handler methods	The callable methods available for the action.
priority	Integer	Controls precedence when multiple actions are eligible.
enabled	true / false	Turns the row on or off.
description	Free text	Explains the action.
created_at	Timestamp	Row creation audit timestamp.

ce_mcp_tool

ce_mcp_tool is the scoped MCP tool registry. In the current v2 line, scope is explicit and mandatory.

ce_mcp_tool columns

Column	Typical values	What it does
tool_id	Integer / PK	Unique tool row id.
tool_code	Unique tool identifier	Primary tool lookup key used by planners and direct tool requests.
tool_group	DB, HTTP_API, WORKFLOW_ACTION, DOCUMENT_RETRIEVAL, CALCULATOR_TRANSFORM, NOTIFICATION, FILES	Normalizes tool execution path and adapter selection.
intent_code	Concrete intent, ANY, UNKNOWN	Mandatory intent scope for the tool.
state_code	Concrete state, ANY, UNKNOWN	Mandatory state scope for the tool.
enabled	true / false	Turns the tool on or off.
description	Free text	Explains what the tool does.
created_at	Timestamp	Row creation audit timestamp.

ce_mcp_db_tool

ce_mcp_db_tool is the SQL-template fallback configuration when the DB tool path is template-driven.

ce_mcp_db_tool columns

Column	Typical values	What it does
tool_id	FK to ce_mcp_tool.tool_id	Binds the DB-specific config to a tool row.
dialect	POSTGRES, SQLITE, ORACLE	Selects the SQL dialect for the template.
sql_template	SQL text	Deterministic SQL template used for execution.
param_schema	JSON schema	Declares the expected input params and their shape.
safe_mode	true / false	Enables the safer execution path for DB SQL templates.
max_rows	Integer	Hard cap on returned rows for this tool template.
created_at	Timestamp	Row creation audit timestamp.
allowed_identifiers	JSON array/object	Optional allowlist for dynamic identifiers in the SQL template.

ce_mcp_sql_guardrail

ce_mcp_sql_guardrail is the shared SQL safety table for MCP DB execution. It keeps the runtime read-only while still letting you explicitly allow or block SQL functions and extra SQL keywords.

ce_mcp_sql_guardrail columns

Column	Typical values	What it does
guardrail_id	Integer / PK	Unique SQL guardrail row id.
rule_type	ALLOW_FUNCTION, BLOCK_FUNCTION, BLOCK_KEYWORD	Determines whether the row extends the function allowlist or blocks a function/keyword.
match_value	date_part, pg_sleep, explain	The SQL function name or keyword matched by the rule.
enabled	true / false	Turns the rule on or off.
description	Free text	Explains why this SQL guardrail row exists.
created_at	Timestamp	Row creation audit timestamp.

ce_mcp_sql_guardrail example values

rule_type	match_value	Runtime effect
ALLOW_FUNCTION	date_part	Adds `date_part(...)` to the allowed function set on top of the built-in safe defaults.
BLOCK_FUNCTION	pg_sleep	Blocks sleep-style functions even though the query is still otherwise read-only.
BLOCK_KEYWORD	explain	Blocks `EXPLAIN` if your policy does not want query-plan inspection through MCP.

Example entry:

INSERT INTO ce_mcp_sql_guardrail (
  rule_type,
  match_value,
  enabled,
  description
) VALUES (
  'BLOCK_FUNCTION',
  'pg_sleep',
  TRUE,
  'Prevent delay or sleep-style SQL calls in all MCP DB execution paths.'
);

Allowed function example:

-- Lets read-only MCP SQL use:
-- select * from fetch_inventory_status(123)
INSERT INTO ce_mcp_sql_guardrail (
  rule_type,
  match_value,
  enabled,
  description
) VALUES (
  'ALLOW_FUNCTION',
  'fetch_inventory_status',
  TRUE,
  'Allow inventory lookup function calls used by read-only MCP diagnostics.'
);

Important:

• the guard only checks that the SQL stays in a read-only SELECT/WITH shape and that fetch_inventory_status is on the allowlist
• it does not inspect the database function body
• if fetch_inventory_status(...) performs writes internally, the SQL layer cannot detect that, so only add ALLOW_FUNCTION rows for functions your DB team guarantees are side-effect free

ce_mcp_planner

ce_mcp_planner makes planner prompts data-driven and scope-aware.

ce_mcp_planner columns

Column	Typical values	What it does
planner_id	Integer / PK	Unique planner row id.
intent_code	Concrete intent, ANY, UNKNOWN	Mandatory intent scope for planner prompt selection.
state_code	Concrete state, ANY, UNKNOWN	Mandatory state scope for planner prompt selection.
system_prompt	Prompt text	System prompt used by the planner.
user_prompt	Prompt text	User prompt template used by the planner.
enabled	true / false	Turns the planner row on or off.
created_at	Timestamp	Row creation audit timestamp.

ce_verbose

ce_verbose is the runtime progress/error message control table added in the later v2 line.

ce_verbose columns

Column	Typical values	What it does
verbose_id	Integer / PK	Unique verbose row id.
intent_code	Concrete intent, ANY, UNKNOWN	Scopes the verbose rule by intent.
state_code	Concrete state, ANY, UNKNOWN	Scopes the verbose rule by state.
step_match	EXACT, REGEX, JSON_PATH	Controls how step_value is matched.
step_value	Step name, regex, or JSON_PATH target	The runtime step selector.
determinant	STEP_ENTER, STEP_EXIT, STEP_ERROR, MCP_TOOL_CALL, RESOLVE_RESPONSE_SELECTED, etc.	The runtime event that triggers the message.
rule_id	Optional integer	Optional rule-specific filter.
tool_code	Optional tool code	Optional tool-specific filter.
message	Verbose text template	Normal progress message.
error_message	Verbose text template	Error-specific message variant.
priority	Integer	Controls winner selection when multiple rows match.
enabled	true / false	Turns the verbose row on or off.
created_at	Timestamp	Row creation audit timestamp.

Summary View

Core static control tables

Primary static ce_* control tables

Table	Primary purpose	Main step(s) that use it	Key fields that matter most
ce_config	Global key/value feature flags and defaults.	Multiple steps and bootstrapping services.	config_type, config_key, config_value, enabled
ce_container_config	Container-position mapping for input params.	AddContainerDataStep.	intent_code, state_code, page_id, section_id, container_id, input_param_name, priority
ce_intent	Intent registry and intent metadata.	IntentResolutionStep, startup cache preload.	intent_code, description, priority, enabled, display_name, llm_hint
ce_intent_classifier	Classifier rules that map user text into intent + initial state.	IntentResolutionStep.	intent_code, state_code, rule_type, pattern, priority, enabled
ce_output_schema	Schema contract for structured extraction and follow-up collection.	SchemaExtractionStep, CorrectionStep, ResponseResolutionStep.	intent_code, state_code, json_schema, priority, enabled
ce_prompt_template	Prompt templates and interaction semantics.	SchemaExtractionStep, ResponseResolutionStep, CorrectionStep, MCP planner prompt selection.	intent_code, state_code, response_type, system_prompt, user_prompt, interaction_mode, interaction_contract
ce_response	Final response selection and derivation behavior.	ResponseResolutionStep.	intent_code, state_code, output_format, response_type, exact_text, derivation_hint, priority
ce_rule	Rule engine transitions and mutations.	RulesStep and phase-specific rule hooks.	phase, intent_code, state_code, rule_type, match_pattern, action, action_value, priority
ce_policy	Pre-intent policy-response rows.	PolicyEnforcementStep.	rule_type, pattern, response_text, priority, enabled
ce_pending_action	Static catalog of confirmable pending tasks.	ActionLifecycleStep, DisambiguationStep, PendingActionStep.	action_key, bean_name, method_names, intent_code, state_code, priority, enabled
ce_mcp_tool	MCP tool registry with strict intent/state scope.	McpToolStep, ToolOrchestrationStep, McpToolRegistry.	tool_code, tool_group, intent_code, state_code, enabled
ce_mcp_db_tool	SQL-template fallback configuration for DB tools.	McpDbToolExecutor / DB tool fallback paths.	tool_id, dialect, sql_template, param_schema, safe_mode, max_rows
ce_mcp_sql_guardrail	Read-only SQL safety extensions for MCP DB paths.	McpSqlGuardrail, McpDbExecutor, SemanticQueryTemplateStepExecutor.	rule_type, match_value, enabled, description
ce_mcp_planner	Planner prompts scoped by intent/state.	McpPlanner.	planner_id, intent_code, state_code, system_prompt, user_prompt, enabled
ce_verbose	Config-driven runtime progress and error messages.	Verbose publishing adapters across many steps.	intent_code, state_code, determinant, step_match, step_value, message, error_message, priority

What these tables actually do

Intent and routing

Intent + routing tables

Table	What it decides	Important values / examples	Why it matters
ce_intent	Which business intents exist.	ORDER_DIAGNOSTICS, LOAN_APPLICATION, SEMANTIC_QUERY_DIAGNOSTICS, GREETING	Every later scoped decision depends on a valid intent.
ce_intent_classifier	How raw text maps to intent/state.	rule_type is pattern-driven and consumer-defined (for example REGEX-style classification rows)	This is the earliest business routing decision.
ce_rule	How the engine mutates state, input params, JSON, tasks, and dialogue act after each phase.	phase values include POST_AGENT_INTENT, POST_SCHEMA_EXTRACTION, PRE_AGENT_MCP, POST_AGENT_MCP, PRE_RESPONSE_RESOLUTION, POST_TOOL_EXECUTION, POST_DIALOGUE_ACT	Most domain behavior should live here instead of hardcoded Java.
ce_policy	Whether a turn should be short-circuited with a policy response before intent work continues.	pattern and response_text form the policy contract	This is the early safety boundary before expensive work starts.

Schema, prompts, and responses

Schema + response tables

Table	What it controls	Important values / examples	Why it matters
ce_output_schema	The structured fields the engine tries to collect or extract.	JSON Schema with required fields and property metadata	Controls missing-field prompting, schema completeness, and correction targeting.
ce_prompt_template	Prompt text and turn semantics.	interaction_mode examples: NORMAL, IDLE, COLLECT, CONFIRM, PROCESSING, FINAL, ERROR, DISAMBIGUATE, FOLLOW_UP, PENDING_ACTION, REVIEW	This now shapes routing semantics as well as prompt text.
ce_response	How the final assistant message is resolved.	output_format examples: TEXT, JSON; response_type examples: EXACT, DERIVED	This remains the final response authority even when MCP produces internal answers.

Actions, tools, and orchestration

Task + MCP tables

Table	What it controls	Important values / examples	Why it matters
ce_pending_action	Which pending tasks are available in a given intent/state.	runtime lifecycle states are OPEN, IN_PROGRESS, EXECUTED, REJECTED, EXPIRED	This keeps user confirmations and task execution table-driven.
ce_mcp_tool	Which tools are visible for a given intent/state.	tool_group values: DB, HTTP_API, WORKFLOW_ACTION, DOCUMENT_RETRIEVAL, CALCULATOR_TRANSFORM, NOTIFICATION, FILES	Prevents cross-intent tool leakage and drives both direct and planner MCP flows.
ce_mcp_db_tool	How DB tools can fall back to deterministic SQL templates.	dialect values such as postgres/sqlite/oracle depending on consumer setup	Used when DB execution is template-driven rather than handler-driven.
ce_mcp_sql_guardrail	How read-only SQL is further constrained or extended.	rule_type values: ALLOW_FUNCTION, BLOCK_FUNCTION, BLOCK_KEYWORD	Lets you stay read-only while still governing which SQL functions and extra keywords are permitted.
ce_mcp_planner	How planner prompts change by domain and state.	resolution order is exact scope -> intent+ANY -> ANY+ANY -> legacy ce_config fallback	This makes MCP planning data-driven instead of globally hardcoded.

Runtime persistence tables

Runtime / audit tables

Table	What it stores	Written by	Why it matters
ce_conversation	Current conversation snapshot and context JSON.	LoadOrCreateConversationStep, PersistConversationStep.	This is the main persisted runtime state.
ce_conversation_history	Turn history snapshots.	Async/history persistence path.	Used for replay, continuity, and follow-up context.
ce_audit	Structured audit trail across pipeline stages.	Almost every step.	This is the first place to inspect when behavior is unexpected.
ce_llm_call_log	Prompt/completion traces for LLM calls.	LLM integration + async persistence path.	Useful when debugging prompt selection and LLM outputs.

Semantic Query metadata tables

Semantic Query ce_semantic_* tables

Table	Purpose	Used by	What it makes possible
ce_semantic_concept	Business concepts (ENTITY/STATUS/INTENT).	SemanticInterpretService.	Maps user language into domain concepts.
ce_semantic_synonym	Synonym table for concept matching.	SemanticInterpretService.	Improves recall for natural-language variants.
ce_semantic_query_class	Query class contracts (filters/sorts/defaults).	SemanticInterpretService, SemanticLlmQueryService.	Constrains query intent and default behavior.
ce_semantic_mapping	Business field -> physical mapping.	SemanticInterpretService, SemanticLlmQueryService.	Enables SQL generation from canonical intent.
ce_semantic_ambiguity_option	Explicit clarification options.	SemanticInterpretService.	Provides deterministic disambiguation choices.
ce_semantic_concept_embedding	Embedding corpus for concepts.	SemanticInterpretService.	Vector-assisted concept/entity retrieval.
ce_semantic_entity	Entity-level override metadata.	SemanticInterpretService.	Patch behavior without Java changes.
ce_semantic_relationship	Relationship override metadata.	SemanticInterpretService.	Override join relationships without Java changes.
ce_semantic_join_hint	Join hints metadata.	SemanticInterpretService.	Steers join planning for complex schemas.
ce_semantic_value_pattern	Value extraction patterns.	SemanticInterpretService.	Improves value parsing from natural language.
ce_semantic_query_failures	Failure memory + corrected SQL + embeddings.	SemanticLlmQueryService, SemanticFailureFeedbackService.	Supports retry, retrieval, and correction learning.

Config families and values

convengine.flow.dialogue-act.*

Dialogue act config

Setting / value	Possible values	What it changes
resolveMode	REGEX_ONLY, REGEX_THEN_LLM, LLM_ONLY	Controls whether dialogue act is cheap regex only or escalates to LLM.
regex patterns	consumer-defined string patterns	Controls fast-path matches for AFFIRM / NEGATE / EDIT / RESET / QUESTION / NEW_REQUEST / ANSWER.
confidence thresholds	numeric	Controls when LLM fallback or confidence-based routing should happen.

convengine.flow.interaction-policy.*

Interaction policy config

Setting / value	Possible values	What it changes
matrix	consumer-defined combinations	Lets you map pending-action / pending-slot / dialogue-act combinations into policy decisions.
executePendingOnAffirm	true / false	AFFIRM can trigger EXECUTE_PENDING_ACTION.
rejectPendingOnNegate	true / false	NEGATE can trigger REJECT_PENDING_ACTION.
fillPendingSlotOnNonNewRequest	true / false	Non-new turns can continue slot filling instead of reclassifying intent.
requireResolvedIntentAndState	true / false	Makes interaction policy depend on already having a stable intent/state.

convengine.flow.action-lifecycle.*

Pending action lifecycle config

Setting / value	Possible values	What it changes
ttlTurns	integer	How many turns a pending action remains alive.
ttlMinutes	integer	How long a pending action remains valid by wall-clock time.
enabled	true / false	Turns ActionLifecycleStep TTL enforcement on or off.

convengine.flow.guardrail.*

Guardrail config

Setting / value	Possible values	What it changes
enabled	true / false	Turns GuardrailStep on or off.
blocked patterns / sanitizers	consumer-defined	Defines what should be blocked or rewritten before later steps run.
approval behavior	consumer-defined	Controls when a turn is allowed, blocked, or requires extra approval handling.

convengine.flow.tool-orchestration.*

Direct tool path config

Setting / value	Possible values	What it changes
enabled	true / false	Turns direct request-driven tool execution on or off.
group policies	consumer-defined	Controls how tool groups are routed when a direct tool_request arrives.

convengine.flow.memory.*

Memory config

Setting / value	Possible values	What it changes
enabled	true / false	Turns MemoryStep session-summary writing on or off.
recentTurnsForSummary	integer	Controls how many recent turns feed memory summarization.
store behavior	consumer-defined implementation choice	Controls where memory is written and how recall is handled.

convengine.flow.query-rewrite.*

Query rewrite config

Setting / value	Possible values	What it changes
enabled	true / false	Turns standalone query rewrite on or off.
rewrite behavior	consumer/LLM-path dependent	Controls when follow-up turns produce standalone_query and resolved_user_input.

convengine.mcp.guardrail.*

MCP next-tool guardrail config

Setting / value	Possible values	What it changes
enabled	true / false	Turns the next-tool guardrail on or off.
failClosed	true / false	When true, missing allowed-next config can block planner proposals.
allowedNextByCurrentTool	map of currentTool -> allowed next tools	Whitelists legal planner transitions between tool calls.

convengine.mcp.http-api.defaults.*

HTTP tool defaults

Setting / value	Possible values	What it changes
timeouts	integer ms values	Controls connect/read limits for framework-managed HTTP calls.
retry	attempt count / backoff values	Controls automatic retry policy.
circuit breaker	threshold / window values	Controls temporary short-circuit behavior for unstable downstreams.
auth hooks	consumer-defined provider config	Controls how outbound auth is attached.

convengine.mcp.db.semantic.*

Semantic Query config

Setting / value	Possible values	What it changes
enabled	true / false	Turns semantic runtime on/off.
toolCode	db.semantic.query	Primary semantic query tool code.
semanticModelSource	DB tables (`ce_semantic_*`)	Semantic model source assembled from DB tables (no YAML path).
defaultLimit / maxLimit	integer	Default and max row limits.
sqlDialect / timezone	string	Dialect and timezone controls for query generation.
retrieval.*	weights + limits	Controls semantic retrieval ranking behavior.
vector.*	enabled/table/columns/maxResults	Controls embedding-based retrieval settings.
clarification.*	threshold + option limits	Controls ambiguity clarification behavior.

High-impact lookup map

If you want to change X, start with Y

What you want to change	Start with this table/config	Why
Intent classification	ce_intent + ce_intent_classifier	That is where raw text becomes an intent/state.
State transitions	ce_rule	Most state changes should stay rule-driven.
Required structured fields	ce_output_schema	That defines extraction and collection requirements.
Prompt text or confirmation semantics	ce_prompt_template	Prompt content and interaction_mode/interaction_contract live there.
Final assistant wording	ce_response	ResponseResolutionStep still resolves the final payload from here.
Pending confirmations or task execution	ce_pending_action + convengine.flow.action-lifecycle.*	Those control the multi-turn task path.
Direct tool availability	ce_mcp_tool + convengine.flow.tool-orchestration.*	That controls tool visibility and direct execution.
Planner behavior	ce_mcp_planner + convengine.mcp.guardrail.*	That controls tool order and whether transitions are allowed.
Semantic intent mapping	ce_semantic_concept + ce_semantic_synonym + ce_semantic_mapping	That controls how user text becomes canonical business intent.
Semantic query behavior	ce_semantic_query_class + ce_semantic_ambiguity_option + ce_semantic_join_hint	That controls query class constraints, clarification, and join guidance.
Semantic retry memory	ce_semantic_query_failures	That controls failure-memory examples and corrected-SQL reuse.
Runtime progress messages	ce_verbose	That is the control-plane source for verbose messages.