Database Schema

The application uses a single SQLite database (evonic.db) with tables for evaluation results, test definitions, and the agent platform.

Evaluation Tables

evaluation_runs

Top-level table for each evaluation run.

Column	Type	Description
run_id	TEXT PK	UUID identifier
started_at	DATETIME	When the run started
completed_at	DATETIME	When the run finished (null if in progress)
model_name	TEXT	Model label for this run
summary	TEXT	Run summary text
overall_score	REAL	Aggregate score (0.0-1.0)
total_tokens	INTEGER	Total tokens consumed
total_duration_ms	INTEGER	Total wall-clock time

test_results

Aggregated results per domain/level.

Column	Type	Description
id	INTEGER PK	Auto-increment
run_id	TEXT FK	References evaluation_runs
domain	TEXT	Domain identifier
level	INTEGER	Complexity level (1-5)
score	REAL	Aggregate score for this cell
status	TEXT	passed, failed, running, pending
prompt	TEXT	Last prompt (legacy)
response	TEXT	Last response (legacy)

individual_test_results

Per-test results with full details.

Column	Type	Description
id	INTEGER PK	Auto-increment
run_id	TEXT FK	References evaluation_runs
test_id	TEXT FK	References tests
domain	TEXT	Domain identifier
level	INTEGER	Level number
prompt	TEXT	Full prompt sent to LLM
response	TEXT	Full LLM response
expected	TEXT (JSON)	Expected output
score	REAL	Score (0.0-1.0)
status	TEXT	passed or failed
details	TEXT (JSON)	Evaluator-specific details
duration_ms	INTEGER	Response time
model_name	TEXT	Model used
system_prompt	TEXT	Resolved system prompt

level_scores

Aggregated scores per domain/level.

Column	Type	Description
run_id	TEXT FK	References evaluation_runs
domain	TEXT	Domain identifier
level	INTEGER	Level number
average_score	REAL	Average score across tests
total_tests	INTEGER	Number of tests
passed_tests	INTEGER	Number of passed tests

---

Test Definition Tables

domains

Cached domain metadata from domain.json files.

Column	Type	Description
id	TEXT PK	Domain identifier
name	TEXT	Display name
evaluator_id	TEXT	Default evaluator
system_prompt	TEXT	Domain system prompt
system_prompt_mode	TEXT	overwrite or append
enabled	BOOLEAN	Include in evaluations
tool_ids	TEXT	JSON array of tool IDs

levels

Per-level configuration.

Column	Type	Description
domain_id	TEXT PK	References domains
level	INTEGER PK	Level number
system_prompt	TEXT	Level system prompt
system_prompt_mode	TEXT	overwrite or append

tests

Individual test definitions.

Column	Type	Description
id	TEXT PK	Test identifier
domain_id	TEXT FK	References domains
level	INTEGER	Level number
prompt	TEXT	User prompt
expected	TEXT (JSON)	Expected output
evaluator_id	TEXT	Override evaluator
system_prompt	TEXT	Test system prompt
weight	REAL	Score weight (default: 1.0)
enabled	BOOLEAN	Include in evaluations

evaluators

Evaluator configuration registry.

Column	Type	Description
id	TEXT PK	Evaluator identifier
type	TEXT	regex, custom, hybrid
eval_prompt	TEXT	LLM evaluation prompt template
extraction_regex	TEXT	Regex pattern for score extraction
uses_pass2	BOOLEAN	Whether to use two-pass extraction
config	TEXT (JSON)	Additional configuration

tools

Tool definition registry.

Column	Type	Description
id	TEXT PK	Tool identifier
name	TEXT	Display name
function_def	TEXT (JSON)	OpenAI function schema
mock_response	TEXT	Mock response for evaluation
mock_response_type	TEXT	json or javascript

---

Agent Platform Tables

agents

Agent definitions.

Column	Type	Description
id	TEXT PK	Slug identifier (e.g., bookstore_bot)
name	TEXT	Display name
description	TEXT	Short description
system_prompt	TEXT	Agent persona and instructions
model	TEXT	Model override (null = use default)
created_at	TIMESTAMP	Creation time
updated_at	TIMESTAMP	Last update time

agent_tools

Many-to-many mapping of agents to tools.

Column	Type	Description
agent_id	TEXT PK, FK	References agents
tool_id	TEXT PK	Tool identifier

channels

Per-agent channel configurations.

Column	Type	Description
id	TEXT PK	UUID identifier
agent_id	TEXT FK	References agents
type	TEXT	telegram, whatsapp, discord
name	TEXT	Display name
config	TEXT (JSON)	Channel-specific config (e.g., bot token)
enabled	BOOLEAN	Whether the channel is active

chat_sessions

Per-user conversation sessions.

Column	Type	Description
id	TEXT PK	UUID identifier
agent_id	TEXT FK	References agents
channel_id	TEXT FK	References channels (nullable for web chat)
external_user_id	TEXT	User identifier from the channel

Sessions are uniquely identified by the tuple (agent_id, channel_id, external_user_id).

chat_messages

Conversation message history.

Column	Type	Description
id	INTEGER PK	Auto-increment
session_id	TEXT FK	References chat_sessions
role	TEXT	user, assistant, tool, system
content	TEXT	Message content
tool_calls	TEXT (JSON)	Tool call objects (for assistant messages)
tool_call_id	TEXT	Tool call ID (for tool result messages)
created_at	TIMESTAMP	Message timestamp