# Export Formats This page documents the file format for each resource export type. All JSON exports include a standard envelope with export_type, software_version, account_id, and exported_at fields. JSONL exports contain one record per line. CSV exports use the same fields as column headers. Export files can be downloaded from the UI or via the [API](https://seclai.com/docs/api-introduction) and [MCP tools](https://seclai.com/docs/mcp). ## Agent Definition (single agent) A standalone, portable snapshot of one agent — workflow, trigger, alerts, evaluation criteria, governance policies, and the resolved dependency manifest. Different in shape from the bulk resource exports below: there's no top-level `export_type` or `account_id` envelope; the snapshot is rooted at the `agent` object with sibling sections for the trigger and extras. This is the format produced by `GET /api/agents/{id}/export` and the `export_agent` MCP tool, and consumed by the matching `agent_definition` field on the create/update endpoints. See the [Agents documentation](https://seclai.com/docs/agents#exporting-agents) for end-to-end usage and the [Importing Agents](https://seclai.com/docs/agents#importing-agents) section for round-trip behavior. **Formats:** JSON only. ### Top-level shape | Field | Type | Description | | ----------------------- | -------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | **export_version** | string | Schema version of the export format. Currently `"2"`. Forward-compatible: imports ignore unknown top-level fields, so older Seclai deployments can usually consume newer exports. | | **exported_at** | string | ISO-8601 timestamp of when the export was generated. | | **software_version** | string | Application version that produced this export. | | **agent** | object | Agent metadata + the full step workflow. See [Agent block](#agent-definition-agent). | | **trigger** | object \| null | Trigger config (input template, content type, schedules). See [Trigger](#agent-definition-trigger). | | **alert_configs** | array \| null | Alert configurations attached to the agent. | | **evaluation_criteria** | array \| null | Per-step evaluation criteria. | | **governance_policies** | array \| null | Agent-scoped governance policies, including KB attachment names. | | **dependencies** | object \| null | Resolved manifest mapping every external UUID referenced in the workflow to a human-readable name (and email for users, source_count for KBs). | | **solutions** | array \| null | Solutions the agent belonged to in the source account. On import, matched by name in the target account; unmatched names are skipped with an `import_warnings` entry. | ### Agent block | Field | Type | Description | | --------------------------------------- | -------------- | ---------------------------------------------------------------------------------------- | | **name** | string | Agent name. | | **description** | string \| null | Agent description. | | **definition** | object | Full workflow definition: `name`, `description`, `tags`, and the recursive `steps` tree. | | **schema_version** | string | Workflow schema version. | | **default_evaluation_tier** | string \| null | `fast`, `balanced`, `thorough`, or null (use account default). | | **evaluation_mode** | string \| null | `output_expectation`, `eval_and_retry`, or `sample_and_flag`. | | **sampling_config** | object \| null | Per-rule sampling configuration when `evaluation_mode = "sample_and_flag"`. | | **max_retries** | int | Retry budget for `eval_and_retry` mode. | | **retry_on_failure** | bool | Whether evaluation failures trigger a retry. | | **prompt_model_auto_upgrade_strategy** | string | `none`, `early_adopter`, `middle_of_road`, or `cautious_adopter`. | | **prompt_model_auto_rollback_enabled** | bool | Whether automatic model rollback is enabled. | | **prompt_model_auto_rollback_triggers** | array \| null | Failure signals that fire rollback (e.g. `agent_eval_fail`, `governance_block`). | | **created_at**, **updated_at** | string | ISO-8601 timestamps. | ### Trigger block | Field | Type | Description | | ---------------------- | -------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | **trigger_type** | string | One of `dynamic_input`, `template_input`, `content_added`, `content_updated`, `agent_run_completed`. | | **name** | string \| null | Display name for template_input triggers. | | **input** | string \| null | Input template (for `template_input`) or default input. | | **input_content_type** | string \| null | MIME type of the input. | | **allow_empty_input** | bool | Whether an empty input is allowed at run time. | | **default_input** | string \| null | Default value when no input is supplied. | | **metadata\_** | object \| null | Free-form metadata attached to the trigger. | | **knowledge_base_id** | string \| null | UUID of the linked KB for content-triggered agents (matched by name on import via `dependencies.knowledge_bases`). | | **schedules** | array \| null | One entry per schedule (only valid on `template_input` triggers). Each: `frequency`, `repeat_interval`, `day_of_week`, `day_of_month`, `hour_of_day`, `minute_of_hour`, `weekday_of_month`. | ### Extras and dependencies The remaining sections (`alert_configs`, `evaluation_criteria`, `governance_policies`, `dependencies`, `solutions`) carry agent extras and the resolved cross-account name lookup table. The full schema is enforced by the import validator at [src/api/services/agent_definition_import.py](https://github.com/anthropics/seclai/blob/main/backend/api/src/api/services/agent_definition_import.py); a 422 response from the create/update endpoints lists any per-field problems with line/column anchors against a canonical pretty-printed echo. **On a cross-account import**, items that don't resolve in the target account are silently skipped (never error) and reported in the response's `import_warnings` array — one entry per skipped item with `category`, `message`, and category-specific `details`. The frontend modal lists these inline; the API and MCP responses include them in the JSON body. ## Governance Policies Exports all governance policies for the account, including scope assignments, thresholds, and linked knowledge bases. **Formats:** JSON, JSONL. **Available filters:** - category — Policy category (e.g. content_safety, pii). - enabled — Only enabled (true) or disabled (false) policies. - account_wide_only — Exclude agent/source-scoped overrides. ### Envelope Fields | Field | Type | Description | | -------------------- | ------------------------------- | -------------------------------------------------------------------- | | **export_type** | string | Resource type identifier (e.g. governance_policies, knowledge_base). | | **software_version** | string | Application version that produced this export. | | **account_id** | string | UUID of the account that owns the exported data. | | **exported_at** | string | ISO-8601 timestamp of when the export was generated. | | **policy_count** | integer | Total number of policies in this export. | | **policies** | array of GovernancePolicyRecord | Array of policy records. | ### Record Fields | Field | Type | Description | | ------------------------ | ----------------------- | ----------------------------------------------------------------------------------------------------- | | **id** | string | UUID of the policy. | | **policy_name** | string | Human-readable policy name. | | **policy_text** | string | Full policy text evaluated by the AI screener. | | **category** | string or null | Policy category (content_safety, pii, bias, legal, brand). | | **enabled** | boolean | Whether the policy is currently active. | | **flag_threshold** | number | Score at or above which content is flagged for review (0.0–1.0). | | **block_threshold** | number | Score at or above which content is blocked (0.0–1.0). | | **review_input** | boolean | Whether this policy screens agent inputs. | | **review_output** | boolean | Whether this policy screens agent/step outputs. | | **enforcement_level** | string or null | How child scopes can modify this policy (flexible, required, locked). | | **inheritance_mode** | string or null | How a scoped policy relates to its parent (inherit, merge, disable). Null for account-level policies. | | **knowledge_base_names** | array of string or null | Names of linked knowledge bases used for evidence-based evaluation. | | **agent_id** | string or null | UUID of the agent this policy is scoped to, or null for account-wide. | | **agent_step_id** | string or null | Agent step ID this policy is scoped to. | | **source_connection_id** | string or null | UUID of the source connection this policy is scoped to. | | **created_at** | string | ISO-8601 creation timestamp. | | **updated_at** | string | ISO-8601 last-update timestamp. | --- ## Governance Evaluations Exports governance evaluation records — the screening results generated when content is checked against governance policies. **Formats:** JSON, JSONL, CSV. **Available filters:** - agent_id — Scope to a specific agent. - source_connection_id — Scope to a specific content source. - policy_id — Filter by governance policy. - verdict — pass, flag, or block. - screening_point — source_content, agent_input, step_output, or policy_test. - unresolved_only — Only unresolved evaluations. ### Envelope Fields | Field | Type | Description | | -------------------- | ----------------------------- | -------------------------------------------------------------------- | | **export_type** | string | Resource type identifier (e.g. governance_policies, knowledge_base). | | **software_version** | string | Application version that produced this export. | | **account_id** | string | UUID of the account that owns the exported data. | | **exported_at** | string | ISO-8601 timestamp of when the export was generated. | | **item_count** | integer | Total number of evaluation records in this export. | | **items** | array of GovernanceEvalRecord | Array of evaluation records. | ### Record Fields | Field | Type | Description | | ------------------------ | -------------- | -------------------------------------------------------------------------------------- | | **id** | string | UUID of the evaluation. | | **policy_id** | string | UUID of the governance policy that was evaluated. | | **policy_name** | string or null | Human-readable name of the policy. | | **screening_point** | string | Where the evaluation occurred (source_content, agent_input, step_output, policy_test). | | **verdict** | string | Evaluation verdict (pass, flag, block). | | **score** | number | AI confidence score (0.0–1.0). | | **flag_threshold_used** | number | Flag threshold applied for this evaluation. | | **block_threshold_used** | number | Block threshold applied for this evaluation. | | **explanation** | string or null | AI-generated explanation of why the content matched (or didn't match) the policy. | | **content_excerpt** | string or null | Excerpt of the evaluated content. | | **agent_id** | string or null | UUID of the agent (if the evaluation was on agent input/output). | | **agent_run_id** | string or null | UUID of the agent run. | | **source_connection_id** | string or null | UUID of the source connection (if the evaluation was on source content). | | **content_version_id** | string or null | UUID of the content version that was evaluated. | | **resolved_at** | string or null | ISO-8601 timestamp when the evaluation was resolved, or null. | | **resolved_by** | string or null | UUID of the user who resolved the evaluation. | | **resolution_note** | string or null | Note left by the reviewer when resolving. | | **created_at** | string | ISO-8601 creation timestamp. | --- ## Knowledge Base Exports a knowledge base with all linked sources and their content items, including full text. **Formats:** JSON only (nested structure). **Available filters:** - from_date — Only content published after this ISO-8601 date. - to_date — Only content published before this ISO-8601 date. ### Envelope Fields | Field | Type | Description | | ----------------------- | ----------------------- | -------------------------------------------------------------------- | | **export_type** | string | Resource type identifier (e.g. governance_policies, knowledge_base). | | **software_version** | string | Application version that produced this export. | | **account_id** | string | UUID of the account that owns the exported data. | | **exported_at** | string | ISO-8601 timestamp of when the export was generated. | | **knowledge_base** | KBMetadata | Knowledge base configuration. | | **source_count** | integer | Number of content sources. | | **total_content_items** | integer | Total content items across all sources. | | **sources** | array of KBSourceRecord | Array of sources with content items. | ### Knowledge base metadata | Field | Type | Description | | --------------------------- | --------------- | ----------------------------------------------- | | **id** | string | UUID of the knowledge base. | | **name** | string | Knowledge base name. | | **description** | string or null | Optional description. | | **reranker_model** | string or null | Reranker model used for search result scoring. | | **default_top_n** | integer or null | Maximum results from the initial vector search. | | **default_top_k** | integer or null | Results kept after reranking. | | **default_score_threshold** | number or null | Minimum relevance score for results (0.0–1.0). | | **created_at** | string or null | ISO-8601 creation timestamp. | | **updated_at** | string or null | ISO-8601 last-update timestamp. | ### Source record | Field | Type | Description | | ------------------------ | ---------------------- | ----------------------------------------- | | **source_connection_id** | string | UUID of the source connection. | | **source_name** | string or null | Human-readable source name. | | **source_url** | string or null | The source's URL (RSS feed or website). | | **polling** | string or null | Polling frequency (e.g. hourly, daily). | | **polling_action** | string or null | What to do on poll (add, replace). | | **polling_max_items** | integer or null | Maximum items to fetch per poll. | | **seed** | string or null | Seed strategy for initial content. | | **retention_days** | integer or null | Content retention period in days. | | **crawl_limit** | integer or null | Maximum pages to crawl (websites only). | | **crawl_depth** | integer or null | Maximum crawl depth (websites only). | | **content_item_count** | integer | Number of content items from this source. | | **content_items** | array of KBContentItem | Array of content items. | ### Content item | Field | Type | Description | | ---------------- | --------------- | ------------------------------------------- | | **id** | string | UUID of the content version. | | **title** | string or null | Content title (e.g. article headline). | | **text** | string or null | Full text content (read from storage). | | **url** | string or null | Source URL of the content item. | | **status** | string or null | Indexing status (indexed, pending, failed). | | **word_count** | integer or null | Number of words in the content. | | **token_count** | integer or null | Number of tokens in the content. | | **published_at** | string or null | ISO-8601 publication timestamp. | | **created_at** | string or null | ISO-8601 creation timestamp. | --- ## Memory Bank Exports a memory bank with its configuration (compaction settings, embedding config) and all entries. **Formats:** JSON only (nested structure). ### Envelope Fields | Field | Type | Description | | -------------------- | ------------------------ | -------------------------------------------------------------------- | | **export_type** | string | Resource type identifier (e.g. governance_policies, knowledge_base). | | **software_version** | string | Application version that produced this export. | | **account_id** | string | UUID of the account that owns the exported data. | | **exported_at** | string | ISO-8601 timestamp of when the export was generated. | | **memory_bank** | MemoryBankMetadata | Memory bank configuration. | | **entry_count** | integer | Total number of entries. | | **entries** | array of MemoryBankEntry | Array of memory bank entries. | ### Memory bank metadata | Field | Type | Description | | --------------------- | --------------- | ------------------------------------------------------------- | | **id** | string | UUID of the memory bank. | | **name** | string | Memory bank name. | | **description** | string or null | Optional description. | | **type** | string or null | Bank type (conversation or general). | | **mode** | string or null | Embedding mode (fast_cheap, balanced, slow_thorough, custom). | | **compaction_prompt** | string or null | Prompt used to summarize entries during compaction. | | **max_age_days** | integer or null | Compaction trigger: compact entries older than this. | | **max_size_tokens** | integer or null | Compaction trigger: compact when total tokens exceed this. | | **max_turns** | integer or null | Compaction trigger: compact when entry count exceeds this. | | **embedding_model** | string or null | Embedding model identifier. | | **dimensions** | integer or null | Vector embedding dimensions. | | **chunk_size** | integer or null | Text chunk size for embedding. | | **chunk_overlap** | integer or null | Overlap between text chunks. | | **created_at** | string or null | ISO-8601 creation timestamp. | | **updated_at** | string or null | ISO-8601 last-update timestamp. | ### Entry record | Field | Type | Description | | -------------------- | -------------- | ----------------------------------------------------------------------------- | | **id** | string | UUID of the content version. | | **title** | string or null | Entry title. | | **text** | string or null | Full text content. | | **published_at** | string or null | ISO-8601 publication timestamp. | | **created_at** | string or null | ISO-8601 creation timestamp. | | **metadata** | object or null | Source-connection metadata (may include key, speaker for conversation banks). | | **content_metadata** | object or null | Content-version metadata. | --- ## Solution Exports a solution with all linked agents, knowledge bases, content sources, and memory banks. Optionally includes full content. **Formats:** JSON only (nested structure). **Available filters:** - include_content — When true, embeds full content items for sources and memory bank entries alongside the resource metadata. Default false (configuration only). ### Envelope Fields | Field | Type | Description | | ---------------------- | --------------------------------- | -------------------------------------------------------------------- | | **export_type** | string | Resource type identifier (e.g. governance_policies, knowledge_base). | | **software_version** | string | Application version that produced this export. | | **account_id** | string | UUID of the account that owns the exported data. | | **exported_at** | string | ISO-8601 timestamp of when the export was generated. | | **solution** | SolutionMetadata | Solution configuration. | | **agents** | array of SolutionAgentRecord | Agents in the solution. | | **knowledge_bases** | array of SolutionKBRecord | Knowledge bases in the solution. | | **source_connections** | array of SolutionSourceRecord | Content sources in the solution. | | **memory_banks** | array of SolutionMemoryBankRecord | Memory banks in the solution. | ### Solution metadata | Field | Type | Description | | --------------- | -------------- | ------------------------------- | | **id** | string | UUID of the solution. | | **name** | string | Solution name. | | **description** | string or null | Description. | | **created_at** | string or null | ISO-8601 creation timestamp. | | **updated_at** | string or null | ISO-8601 last-update timestamp. | ### Agent record | Field | Type | Description | | --------------------------- | --------------- | ----------------------------------- | | **id** | string | UUID of the agent. | | **name** | string | Agent name. | | **description** | string or null | Agent description. | | **evaluation_mode** | string or null | Evaluation mode (if configured). | | **default_evaluation_tier** | string or null | Default governance evaluation tier. | | **max_retries** | integer or null | Maximum evaluation retries. | | **created_at** | string or null | ISO-8601 creation timestamp. | | **updated_at** | string or null | ISO-8601 last-update timestamp. | ### Knowledge base record | Field | Type | Description | | --------------------------- | --------------- | ---------------------------------- | | **id** | string | UUID of the knowledge base. | | **name** | string | Knowledge base name. | | **description** | string or null | Description. | | **reranker_model** | string or null | Reranker model. | | **default_top_n** | integer or null | Top N initial search results. | | **default_top_k** | integer or null | Top K after reranking. | | **default_score_threshold** | number or null | Minimum relevance score (0.0–1.0). | | **created_at** | string or null | ISO-8601 creation timestamp. | | **updated_at** | string or null | ISO-8601 last-update timestamp. | ### Source connection record | Field | Type | Description | | --------------------- | ----------------------- | ---------------------------------------------------------------------------------------------------- | | **id** | string | UUID of the source connection. | | **name** | string or null | Source name. | | **source_url** | string or null | Source URL. | | **polling** | string or null | Polling frequency. | | **polling_action** | string or null | Polling action. | | **polling_max_items** | integer or null | Maximum items per poll. | | **seed** | string or null | Seed strategy. | | **retention_days** | integer or null | Retention period in days. | | **crawl_limit** | integer or null | Maximum pages to crawl. | | **crawl_depth** | integer or null | Maximum crawl depth. | | **created_at** | string or null | ISO-8601 creation timestamp. | | **updated_at** | string or null | ISO-8601 last-update timestamp. | | **content_items** | array of object or null | Content items (only when include_content is true). Each item has id, title, text, url, published_at. | ### Memory bank record | Field | Type | Description | | --------------------- | ----------------------- | ----------------------------------------------------------------------------------------------------------------------------- | | **id** | string | UUID of the memory bank. | | **name** | string | Memory bank name. | | **description** | string or null | Description. | | **type** | string or null | Bank type (conversation or general). | | **mode** | string or null | Embedding mode. | | **compaction_prompt** | string or null | Compaction prompt. | | **max_age_days** | integer or null | Compaction trigger: max age in days. | | **max_size_tokens** | integer or null | Compaction trigger: max tokens. | | **max_turns** | integer or null | Compaction trigger: max turns. | | **embedding_model** | string or null | Embedding model. | | **dimensions** | integer or null | Vector dimensions. | | **chunk_size** | integer or null | Chunk size. | | **chunk_overlap** | integer or null | Chunk overlap. | | **created_at** | string or null | ISO-8601 creation timestamp. | | **updated_at** | string or null | ISO-8601 last-update timestamp. | | **entries** | array of object or null | Memory bank entries (only when include_content is true). Each entry has id, title, text, published_at, and optional metadata. | --- ## Agent Traces Exports agent run traces with step-level execution detail, including inputs, outputs, attempts, timing, and error information. **Formats:** JSON, JSONL. **Available filters:** - status — Filter runs by status (e.g. completed, failed). - limit — Maximum number of runs to export. - include_step_io — Include step input/output content (default true). ### Envelope Fields | Field | Type | Description | | -------------------- | ------------------------- | -------------------------------------------------------------------- | | **export_type** | string | Resource type identifier (e.g. governance_policies, knowledge_base). | | **software_version** | string | Application version that produced this export. | | **account_id** | string | UUID of the account that owns the exported data. | | **exported_at** | string | ISO-8601 timestamp of when the export was generated. | | **item_count** | integer | Total number of agent run traces. | | **items** | array of AgentTraceRecord | Array of agent run traces. | ### Record Fields | Field | Type | Description | | --------------- | --------------------------- | ---------------------------------------------------- | | **id** | string | UUID of the agent run. | | **status** | string or null | Run status (completed, failed, pending, processing). | | **error_count** | integer or null | Total errors across all steps. | | **priority** | integer or null | Run priority. | | **call_depth** | integer or null | Nesting depth for sub-agent calls. | | **created_at** | string or null | ISO-8601 creation timestamp. | | **updated_at** | string or null | ISO-8601 last-update timestamp. | | **step_count** | integer | Number of step runs. | | **steps** | array of AgentStepRunRecord | Step-level execution records. | | **usages** | array of object | Credit usage records for this run. | ### Step run record | Field | Type | Description | | ----------------------- | ------------------------- | --------------------------------------------------- | | **id** | string | UUID of the step run. | | **step_id** | string | Step identifier within the agent definition. | | **step_type** | string or null | Step type (e.g. prompt_call, retrieval, gate). | | **status** | string or null | Step run status (completed, failed, pending). | | **error_count** | integer or null | Number of errors encountered. | | **input_checksum** | string or null | Checksum of the step input. | | **output_content_type** | string or null | MIME type of the step output. | | **created_at** | string or null | ISO-8601 creation timestamp. | | **attempts** | array of AgentStepAttempt | Execution attempts for this step. | | **input** | string or null | Step input content (when include_step_io is true). | | **output** | string or null | Step output content (when include_step_io is true). | ### Step attempt record | Field | Type | Description | | --------------- | -------------- | ----------------------------------- | | **id** | string | UUID of the attempt. | | **status** | string or null | Attempt status (completed, failed). | | **started_at** | string or null | ISO-8601 start timestamp. | | **ended_at** | string or null | ISO-8601 end timestamp. | | **duration** | number or null | Duration in seconds. | | **error** | string or null | Error message (if failed). | | **cache_stats** | object or null | Cache statistics for this attempt. | --- ## Agent Evaluations Exports agent evaluation results with criteria metadata. Each record joins a criteria definition to its evaluation result. **Formats:** JSON, JSONL, CSV. **Available filters:** - criteria_id — Export results for a single criteria. - status — Filter results by status (passed, failed, error). ### Envelope Fields | Field | Type | Description | | -------------------- | ------------------------ | -------------------------------------------------------------------- | | **export_type** | string | Resource type identifier (e.g. governance_policies, knowledge_base). | | **software_version** | string | Application version that produced this export. | | **account_id** | string | UUID of the account that owns the exported data. | | **exported_at** | string | ISO-8601 timestamp of when the export was generated. | | **item_count** | integer | Total number of evaluation results. | | **items** | array of AgentEvalRecord | Array of evaluation results. | ### Record Fields | Field | Type | Description | | --------------------------- | --------------- | ---------------------------------------------------------------------- | | **criteria_id** | string | UUID of the evaluation criteria. | | **criteria_step_id** | string | Agent step ID targeted by the criteria. | | **criteria_mode** | string or null | Evaluation mode (output_expectation, eval_and_retry, sample_and_flag). | | **criteria_enabled** | boolean | Whether the criteria is currently enabled. | | **criteria_pass_threshold** | number or null | Minimum score to pass (0.0–1.0). | | **result_id** | string | UUID of the evaluation result. | | **agent_run_id** | string | UUID of the agent run that was evaluated. | | **agent_step_run_id** | string or null | UUID of the specific step run. | | **status** | string or null | Result status (passed, failed, pending, skipped, error). | | **score** | number or null | Evaluation score (0.0–1.0). | | **details** | object or null | Evaluation-specific details (JSON). | | **retry_triggered** | boolean or null | Whether this result triggered a retry. | | **retry_count** | integer or null | Number of retries attempted so far. | | **flagged** | boolean or null | Whether the run was flagged for review. | | **evaluated_at** | string or null | ISO-8601 timestamp of the evaluation. | | **created_at** | string or null | ISO-8601 creation timestamp. | --- ## Common Envelope Fields Every JSON export file includes these top-level fields: | Field | Type | Description | | -------------------- | ------ | -------------------------------------------------------------------- | | **export_type** | string | Resource type identifier (e.g. governance_policies, knowledge_base). | | **software_version** | string | Application version that produced this export. | | **account_id** | string | UUID of the account that owns the exported data. | | **exported_at** | string | ISO-8601 timestamp of when the export was generated. | JSONL and CSV exports do not include envelope fields — each line or row contains only record-level data.