> ## Documentation Index > Fetch the complete documentation index at: https://docs.pinecone.io/llms.txt > Use this file to discover all available pages before exploring further. # Get a task (with steps) > Unlike the listing, this carries the task's `steps` — every one by default, or the newest N with `steps_limit`, where `steps_total` reports the true count. ## OpenAPI ````yaml https://raw.githubusercontent.com/pinecone-io/pinecone-api/refs/heads/main/2026-07/nexus_data_2026-07.oas.yaml get /tasks/{id} openapi: 3.0.3 info: title: Nexus API description: > Nexus turns a set of sources into a queryable, self-improving knowledge base. A context is **sources + a manifest**: upload sources, **curate** them into a searchable index, then query it. The manifest, one validated JSON document, defines how the context indexes, retrieves and answers. Curate is explicit — nothing is queryable until you run it. `{slug}` accepts a context's slug or its UUID. Query with `POST /api/query`, one turn per call, and read the answer from `output[].content[].text`; multi-turn conversations are sessions. Tasks are project-owned at `/api/tasks`, not nested under contexts. contact: name: Pinecone Support url: https://support.pinecone.io email: support@pinecone.io license: name: Apache 2.0 url: https://www.apache.org/licenses/LICENSE-2.0 version: 2026-07 servers: - url: https://{host}/api description: Your Nexus deployment host variables: host: default: api.example.com description: Host of your Nexus deployment. security: - bearerAuth: [] - apiKey: [] tags: - name: Auth description: Login and identity. - name: Project description: The active Pinecone project and its disclosure state. - name: Contexts description: A context is sources + a manifest. Curate it explicitly before querying. - name: Manifest description: Manifest templates and the per-context manifest. - name: Source Files description: >- Per-context source imports (upload, connector, public repo) and the source file tree. - name: Knowledge description: Read-only browse of curated knowledge (chunks + artifacts). - name: Curation description: Curate workflow trigger, curation ledger, and version-pin primitives. - name: Connectors description: >- Project-level links to external source providers (Box, ...). Linked via OAuth or an API key, then used to import source documents into a context. - name: Query description: >- The unified KnowQL Query API: run a query turn (`/query`), fetch a turn (`/queries/{id}`), inspect its trace, manage feedback, comparison flags, and multi-turn sessions (`/sessions`). - name: Tasks description: >- Top-level, project-owned task records for every workflow (optimize, curate, search, work, explore, profile, import, pack, restore, groom). - name: Task Files description: Read, list, and delete files from live or archived task containers. paths: /tasks/{id}: get: tags: - Tasks summary: Get a task (with steps) description: >- Unlike the listing, this carries the task's `steps` — every one by default, or the newest N with `steps_limit`, where `steps_total` reports the true count. operationId: get_task parameters: - in: path name: id description: Task id. required: true schema: type: string style: simple - in: header name: X-Pinecone-Api-Version description: >- Date-based contract version, echoed back on the same header. Omit for the default (`2026-07`); send `unstable` for the in-development surface. An unrecognized value is rejected with `400 unsupported_api_version`. schema: default: 2026-07 x-enum: - 2026-07 - unstable type: string style: simple - in: query name: steps_limit description: Return only this many trailing steps. Absent returns them all. schema: type: integer minimum: 0 style: form responses: '200': description: The task content: application/json: schema: $ref: '#/components/schemas/TaskResponse' '404': description: Not found content: application/json: schema: $ref: '#/components/schemas/Error' components: schemas: TaskResponse: description: One task record — a single run of a workflow, owned by the project. type: object properties: id: description: Task id. type: string project_id: description: The Pinecone project that owns the task. type: string context_id: nullable: true description: >- The context the run acts on. Null for a task that belongs to no context. type: string agent_id: nullable: true description: Reserved. Null on every task this version serves. type: string session_id: nullable: true description: Set for query turns type: string created_by: description: Principal that started the run. type: string workflow: $ref: '#/components/schemas/Workflow' state: $ref: '#/components/schemas/TaskState' error: nullable: true description: Failure detail. Set on a `failed` task. type: string input: $ref: '#/components/schemas/TaskInput' output: $ref: '#/components/schemas/TaskOutput' steps: description: >- Populated only on GET /tasks/{id}. May be a trailing window when `steps_limit` was given, so read the count from `steps_total` rather than this array's length. type: array items: $ref: '#/components/schemas/TaskStep' steps_total: description: >- True number of recorded steps. Absent on a read that carries no steps. type: integer format: int64 tokens_prompt: description: Input tokens the run has billed so far. type: integer format: int64 tokens_completion: description: Output tokens the run has billed so far. type: integer format: int64 runtime_seconds: description: Seconds the container has been running. type: integer format: int64 running_from: nullable: true description: When the container started. Null before it does. type: string format: date-time timeout_seconds: nullable: true description: Runtime budget for this task. Null when it runs uncapped. type: integer timeout_at: nullable: true description: When the budget expires and the run is terminated. type: string format: date-time archived_at: nullable: true description: When the task's files were archived. Null while they are still live. type: string format: date-time created_at: description: When the task row was created. type: string format: date-time last_activity_at: nullable: true description: When the runtime last reported anything. Drives stall detection. type: string format: date-time schedule: nullable: true description: >- Cron expression this run was scheduled from. Null for an on-demand run. type: string scheduled_at: nullable: true description: When a `scheduled` task is due to start. type: string format: date-time parent_task_id: nullable: true description: The task that spawned this one, as an optimize spawns its curates. type: string progress: $ref: '#/components/schemas/TaskProgress' required: - id - project_id - created_by - workflow - state - input - steps - tokens_prompt - tokens_completion - runtime_seconds - created_at Error: description: The error envelope every non-2xx response uses. type: object properties: message: description: >- Human-readable explanation. Not stable — branch on `code`, not on this. type: string code: description: >- Stable machine-readable code, present only on errors a client must branch on. x-enum: - unsupported_api_version - preview_not_enabled - preview_file_limit - preview_byte_limit - workspace_host_required - workspace_not_found type: string required: - message Workflow: description: The canonical workflow name. x-enum: - optimize - curate - search - search_cc - search_rag - work - explore - profile - import - pack - restore - groom type: string TaskState: description: >- Where the run has got to. `scheduled` waits for its due time; `starting` and `provisioning` are a container being claimed and built; `running` is the work happening; `stopping` is a termination in progress. `completed`, `cancelled` and `failed` are terminal. x-enum: - scheduled - starting - provisioning - running - stopping - completed - cancelled - failed type: string TaskInput: description: >- A task's per-workflow `input`. It carries no discriminant of its own — the sibling `workflow` field selects the shape. anyOf: - $ref: '#/components/schemas/SearchInput' - $ref: '#/components/schemas/CurateInput' - $ref: '#/components/schemas/ImportInput' - $ref: '#/components/schemas/PackInput' - $ref: '#/components/schemas/RestoreInput' - $ref: '#/components/schemas/ExploreInput' - $ref: '#/components/schemas/ProfileInput' - $ref: '#/components/schemas/OptimizeInput' - $ref: '#/components/schemas/GroomInput' TaskOutput: nullable: true description: >- A task's reported `output`. Every workflow reports a different per-state shape, matched to a named variant only when its key set is exactly that variant's; anything else is preserved as stored. Treat this union as open. Null until the run reports anything. anyOf: - $ref: '#/components/schemas/SearchTurnCompletedOutput' - $ref: '#/components/schemas/SearchCcTurnCompletedOutput' - $ref: '#/components/schemas/SearchTurnFailedOutput' - $ref: '#/components/schemas/PackRunningOutput' - $ref: '#/components/schemas/PackCompletedOutput' - $ref: '#/components/schemas/RestoreRunningOutput' - $ref: '#/components/schemas/RestoreCompletedOutput' - $ref: '#/components/schemas/OptimizeResultOutput' - $ref: '#/components/schemas/OptimizeNoopOutput' - $ref: '#/components/schemas/GroomMetricsOutput' - $ref: '#/components/schemas/CurateRunOutput' - $ref: '#/components/schemas/CurateManifestTuneOutput' - $ref: '#/components/schemas/CurateNoopOutput' - $ref: '#/components/schemas/ExploreOutput' - $ref: '#/components/schemas/ProfileEstimateOutput' - $ref: '#/components/schemas/ProfileEmptyOutput' - $ref: '#/components/schemas/ImportOutput' - $ref: '#/components/schemas/StatusErrorOutput' - $ref: '#/components/schemas/StateFailedOutput' - $ref: '#/components/schemas/ErrorOutput' - title: Other description: Any shape without a named variant type: object additionalProperties: true TaskStep: description: >- One reported step row. `step_id` and `status` are always set; the rest are per-workflow optionals, so an absent key stays absent. type: object properties: step_id: description: Identifies the step within the task. type: string status: description: How the step ended, e.g. `running` or `completed`. type: string type: description: What kind of step this is; the vocabulary is per-workflow. type: string content: description: The step's reported text payload. type: string commentary: description: The runtime's one-line account of what the step is doing. type: string code: description: Redacted tool input (search) type: string result: description: Redacted tool output (search) type: string cum_input_tokens: description: Task-to-date input tokens, including this step. type: integer format: int64 cum_output_tokens: description: Task-to-date output tokens, including this step. type: integer format: int64 job_id: description: Identifier of the LlamaParse job doing the work. type: string path: description: Source path being parsed type: string required: - step_id - status TaskProgress: description: >- Coarse run progress, written by the runtime. Absent on a task that reports none. type: object properties: phase: description: 1-based index of the phase now running. type: integer format: int64 phases: description: Total phases this run expects. type: integer format: int64 label: description: What the current phase is doing. type: string pct: description: Completion across the whole run, 0–100. type: number format: double eta_seconds: nullable: true description: Estimated seconds remaining. Null when the run cannot estimate one. type: integer format: int64 required: - phase - phases - label - pct SearchInput: title: Search / search_cc / search_rag / work description: >- One shape for the whole search family — the same payload is written whichever of them a turn resolves to. type: object properties: session_id: description: The session the turn runs in. type: string query_id: description: The turn this task serves. type: string ask: description: >- Title fallback for the tasks UI; empty when the turn had no user message type: string messages: description: The conversation history handed to the runtime. type: array items: description: One message of history. type: object properties: role: description: Who sent it, e.g. `user`. type: string content: description: The message text. type: string required: - role - content models: description: The session's model fallback list, in preference order. type: array items: description: A catalog model id or tier name. type: string tools: description: Tool names the turn may call. type: array items: description: A tool name. type: string shape: nullable: true description: JSON Schema the answer must conform to. Null on a free-text turn. type: object additionalProperties: true instructions: nullable: true description: The session's system prompt. Null when none was pinned. type: string scope: description: Context ids the turn searches. type: array items: description: A context id the turn searched. type: string turn_timeout_seconds: description: Runtime budget for this one turn. type: integer format: int64 retrieval: description: The turn's retrieval controls. type: object properties: retrieval_only: nullable: true description: >- Skip synthesis and return the hits. Null where the caller omitted it. type: boolean pointers_only: nullable: true description: >- Skip synthesis and return pointers only. Null where the caller omitted it. type: boolean chunks_only: nullable: true description: Narrow retrieval to chunks. Null where the caller omitted it. type: boolean artifacts_only: nullable: true description: Narrow retrieval to artifacts. Null where the caller omitted it. type: boolean max_retrieved: nullable: true description: Cap on retrieved items. Null where the caller omitted it. type: integer format: int64 max_retrieved_chars: nullable: true description: Cap on per-item text length. Null where the caller omitted it. type: integer format: int64 compose: nullable: true description: >- Whether to synthesize an answer. Null where the caller omitted it. type: boolean max_steps: nullable: true description: >- Cap on the agent's tool-loop steps. Null where the caller omitted it. type: integer format: int64 thinking_level: nullable: true description: Reasoning depth. Null where the caller omitted it. x-enum: - minimal - low - medium - high type: string required: - retrieval_only - pointers_only - chunks_only - artifacts_only - max_retrieved - max_retrieved_chars - compose - max_steps - thinking_level comparison_group: nullable: true description: The Compare run this turn belongs to. Null on a normal single query. type: string required: - session_id - query_id - ask - messages - models - tools - shape - instructions - scope - turn_timeout_seconds - comparison_group CurateInput: title: Curate description: What a curate run was started with. type: object properties: trigger: description: >- What started the run, e.g. `manual`, `reconcile`, `restore`, or `resume`. type: string force: description: Full rebuild rather than incremental. type: boolean scope: description: What the run covered, e.g. `full`, `reconcile`, or `restore`. type: string batch_size: description: Sources processed per batch. type: integer models: description: Model-id overrides for the curate runtime. type: array items: description: A catalog model id or tier name. type: string legs: description: >- Set to `["chunks"]` to gate the run to the chunk + keyword indices, skipping artifact extraction. type: array items: description: A leg name, e.g. `chunks`. type: string artifacts_reset: description: Drop the existing artifacts and re-extract from scratch. type: boolean skip_corpus_aggregation: description: >- Skip the corpus-wide artifact pass, leaving only per-document extraction. type: boolean claimed_version: description: What a drained predecessor had already reserved. type: integer format: int64 required: - trigger - force - scope ImportInput: title: Import description: >- One of three disjoint shapes, discriminated by which of `connector_id` / `url` / `raw_archive_key` is present. anyOf: - title: Connector import description: Sources pulled from a linked project connector. type: object properties: provider: description: The connector's provider, e.g. `box`. type: string connector_id: description: The linked connector the sources came from. type: string folder: nullable: true description: Connector folder imported. Null on a file-list import. type: string item_type: description: Whether `folder` or `files` selected the sources. x-enum: - folder - file type: string files: description: Connector file ids imported. Empty on a folder import. type: array items: description: A connector file id. type: string path: description: Subdirectory under the source tree the import landed in. type: string limits: $ref: '#/components/schemas/ImportLimits' required: - provider - connector_id - item_type - files - path - title: Public-repo import description: Sources fetched from a public URL. type: object properties: provider: description: The public-source provider, e.g. `github`. type: string url: description: The public repository or archive fetched. type: string path: description: Subdirectory under the source tree the import landed in. type: string limits: $ref: '#/components/schemas/ImportLimits' required: - provider - url - path - title: Archive upload description: Sources expanded from an uploaded archive. type: object properties: raw_archive_key: description: Staging key under the source bucket's `_inbox/` prefix type: string path: description: Subdirectory under the source tree the import landed in. type: string limits: $ref: '#/components/schemas/ImportLimits' required: - raw_archive_key - path PackInput: title: Pack description: What a pack run was started with. type: object properties: include_traces: description: Whether query traces were packed alongside sources and knowledge. type: boolean slug: description: Slug of the packed context; names the output archive. type: string context_id: description: The context that was packed. type: string required: - include_traces - slug - context_id RestoreInput: title: Restore description: What a restore run was started with. type: object properties: pack_key: description: Object-store key of the pack being restored. type: string pack_filename: description: The pack archive's file name. type: string target_context_id: description: The already-created context the pack is restored into. type: string target_slug: description: URL-safe name of the context being written into. type: string target_name: description: Human-readable label of the context being written into. type: string sample_queries: description: >- Sample queries to seed the target with, taken from the pack. Absent on a pack that carried none. type: array items: description: One sample query carried over from the pack. type: string required: - pack_key - pack_filename - target_context_id - target_slug - target_name ExploreInput: title: Explore description: What an explore run was started with. type: object properties: guide: description: Guidance the proposal was steered with. Empty when none was given. type: string trigger: description: What started the run, e.g. `manual`. type: string required: - guide - trigger ProfileInput: title: Profile description: What a profile run was started with. type: object properties: manifest: description: The candidate manifest the run profiles, not a stored one. allOf: - $ref: '#/components/schemas/Manifest' trigger: description: What started the run, e.g. `manual`. type: string required: - manifest - trigger OptimizeInput: title: Optimize description: >- Only the keys the trigger request supplied are written, so an input with no keys is a valid no-op optimize. type: object properties: candidate_queries: description: The queries the run tunes toward. Absent makes the run a no-op. type: array items: $ref: '#/components/schemas/CandidateQuery' eval_pass_rate_threshold: description: Pass rate an iteration must clear to count as ready. type: number format: double retrieval_p90_latency_ms: description: The p90 latency bar an iteration must stay under. type: integer format: int64 optimize_timeout_seconds: description: Runtime budget for the whole optimize. type: integer format: int64 max_tool_turns: description: Cap on tool-loop turns per eval query. type: integer max_iterations: description: Cap on candidate manifests tried. type: integer models: description: Tier names or catalog ids, in preference order. type: array items: description: A tier name or catalog id. type: string GroomInput: title: Groom description: What a groom run was started with. type: object properties: budget: description: Cap on fact batches the run consolidates. type: integer groom_timeout_seconds: description: Runtime budget for the whole groom. type: integer format: int64 models: description: Tier names or catalog ids, in preference order. type: array items: description: A tier name or catalog id. type: string SearchTurnCompletedOutput: title: Search / search_rag completed description: >- What a finished search turn's task reports. The turn itself, with its steps and structured citations, is at `GET /queries/{query_id}`. type: object properties: status: description: How the turn ended. type: string query_id: description: The turn this task answered. type: string answer: description: The answer text. type: string output_json: nullable: true description: >- Structured answer when the turn carried a shape, or the hits on a retrieval-only turn. Absent otherwise. type: object additionalProperties: true citations: description: Cited source paths. The turn carries the structured form. type: array items: description: A cited source path. type: string latency_ms: description: Wall time for the turn. type: integer format: int64 unknown_models: description: >- Model selections dropped as unknown before dispatch, so a caller whose pin was silently ignored can see it. Absent when nothing was dropped. type: array items: description: >- A requested model id or tier name the project's catalog does not carry. type: string required: - status - query_id - answer - citations - latency_ms SearchCcTurnCompletedOutput: title: Search-as-code completed description: >- The search-as-code path reports a step count where the others report `output_json`. type: object properties: status: description: How the turn ended. type: string query_id: description: The turn this task answered. type: string answer: description: The answer text. type: string citations: description: Cited source paths. type: array items: description: A cited source path. type: string latency_ms: description: Wall time for the turn. type: integer format: int64 steps: description: Reasoning steps the turn ran. type: integer format: int64 required: - status - query_id - answer - citations - latency_ms - steps SearchTurnFailedOutput: title: Search failed description: What a search turn reports when it could not answer. type: object properties: status: description: How the turn ended. type: string query_id: description: The turn that failed. type: string error: description: Why the turn failed. type: string required: - status - query_id - error PackRunningOutput: title: Pack running description: >- A pack's progress: the initial claim, the enrichment pass, then periodic write progress, each adding to what came before. `state` is `running` on all of them, so key presence is what tells them apart. type: object properties: state: description: >- Always `running`. Which report this is shows in the keys it carries, not here. x-enum: - running type: string slug: description: Slug of the context being packed. type: string include_traces: description: >- Whether query traces are being packed. Absent on the write-progress report. type: boolean started_at: description: When the pack started. Absent on the write-progress report. type: string sources_total: description: Source files to be packed. Absent until the inventory is counted. type: integer format: int64 knowledge_total: description: >- Curated knowledge files to be packed. Absent until the inventory is counted. type: integer format: int64 traces_total: description: Trace blobs to be packed. Absent until the inventory is counted. type: integer format: int64 bytes_collected: description: Bytes read so far. Absent until the inventory is counted. type: integer format: int64 enriching: description: Present while the LLM writes the pack's summary. type: boolean files_written: description: >- Files written into the archive so far. Present only once writing has begun. type: integer format: int64 required: - state - slug PackCompletedOutput: title: Pack completed description: >- The archive is written. Download it from `GET /tasks/{id}/files/read/{pack_filename}`. type: object properties: state: description: Which stage of the pack this report describes. type: string pack_filename: description: Name of the written archive, `.context.zip`. type: string pack_size: description: Byte size of the archive. type: integer format: int64 sources_files: description: How many corpus documents went into the archive. type: integer format: int64 knowledge_files: description: How many curated documents went into the archive. type: integer format: int64 traces_included: description: Whether query traces were included. type: boolean traces_files: description: Trace blobs packed. type: integer format: int64 corpus_groups_included: description: Corpus groups carried over from the curation ledger. type: integer format: int64 started_at: description: When the pack started. type: string completed_at: description: When the pack finished. type: string details: $ref: '#/components/schemas/PackDetails' required: - state - pack_filename - pack_size - sources_files - knowledge_files - traces_included - traces_files - corpus_groups_included - started_at - completed_at - details RestoreRunningOutput: title: Restore running description: >- A restore's progress: the initial claim, the fetch, the unzip, then periodic staging. `state` is `running` on all four, so key presence is what tells them apart, and only `state` is on every one. type: object properties: state: description: >- Always `running`. Which report this is shows in the keys it carries, not here. x-enum: - running type: string pack_filename: description: The pack archive's file name. Absent on the staging report. type: string pack_size: description: Byte size of the pack archive. Present once it has been fetched. type: integer format: int64 target_context_id: description: The context the pack is restored into. Absent on the staging report. type: string target_slug: description: >- Slug of that target context. Absent from the unzip and staging reports. type: string started_at: description: When the restore started. Present on the first report only. type: string stage_progress: description: >- How far the current phase has got. Absent on the first report, before any phase begins. allOf: - $ref: '#/components/schemas/StageProgress' sources_count: description: Source files found in the pack. Present once it has been expanded. type: integer format: int64 knowledge_count: description: >- Knowledge files found in the pack. Present once it has been expanded. type: integer format: int64 required: - state RestoreCompletedOutput: title: Restore completed description: >- The target holds the pack's sources, knowledge, and manifest. Its index stays empty until curated. type: object properties: state: description: Which stage of the restore this report describes. type: string pack_filename: description: The pack archive's file name. type: string pack_size: description: How large the archive is on disk. type: integer format: int64 target_context_id: description: The context the pack is restored into. type: string target_slug: description: URL-safe name of the context being written into. type: string sources_staged: description: Source files written into the target. type: integer format: int64 knowledge_staged: description: Knowledge files written into the target. type: integer format: int64 manifest_set: description: Whether the pack's manifest was pinned onto the target. type: boolean corpus_groups_set: description: Corpus groups written into the target's curation ledger. type: integer format: int64 sample_queries: description: >- Sample queries carried from the pack onto the target. Empty when the pack had none. type: array items: description: One sample query set on the target. type: string started_at: description: When the restore started. type: string completed_at: description: When the restore finished. type: string required: - state - pack_filename - pack_size - target_context_id - target_slug - sources_staged - knowledge_staged - manifest_set - corpus_groups_set - sample_queries - started_at - completed_at OptimizeResultOutput: title: Optimize result description: What the tuning loop settled on. type: object properties: status: description: >- Set to `ready` only when the best iteration cleared both the pass-rate and latency bars. type: string best_score: description: Eval pass rate the best iteration scored, 0–1. type: number format: double best_iteration: description: Which iteration scored it. type: integer format: int64 best_latency_ok: description: Whether that iteration also stayed under the latency bar. type: boolean total_iterations: description: Candidate manifests tried. type: integer format: int64 pass_rate_threshold: description: The pass-rate bar the run measured against. type: number format: double p90_latency_limit_ms: nullable: true description: The p90 latency bar. Null when latency was not gated. type: number format: double elapsed_seconds: description: Wall time for the whole loop. type: number format: double usage: $ref: '#/components/schemas/ChatUsage' models: description: Tier → resolved model id. type: object additionalProperties: description: The model id the tier resolved to. type: string unknown_models: description: >- Model selections dropped as unknown before dispatch, so a caller whose pin was silently ignored can see it. Absent when nothing was dropped. type: array items: description: >- A requested model id or tier name the project's catalog does not carry. type: string manifest: description: >- The tuned manifest persisted onto the context when the task completes. allOf: - $ref: '#/components/schemas/Manifest' changed_legs: description: >- Which manifest legs the tuning changed, so a re-curate can be scoped to them. type: array items: description: A manifest leg the tuning changed. type: string required: - status - best_score - best_iteration - best_latency_ok - total_iterations - pass_rate_threshold - p90_latency_limit_ms - elapsed_seconds - usage - models - manifest - changed_legs OptimizeNoopOutput: title: Optimize no-op description: Too few candidate queries or problems to tune from. type: object properties: status: description: How the run ended. type: string reason: description: Which gate the run stopped at. type: string candidate_queries: description: Candidate queries the run was given. type: integer format: int64 problems: description: How many of those counted as problems worth tuning for. type: integer format: int64 required: - status - reason - candidate_queries - problems GroomMetricsOutput: title: Groom description: >- What a groom run consolidated. Groom is to a work context what curate is to a search context — it distills facts out of accumulated work traces into artifacts. type: object properties: status: description: How the run ended. type: string budget: description: Fact batches the run was allowed to consume. type: integer format: int64 batches_staged: description: Batches available to consume. type: integer format: int64 batches_consumed: description: Batches actually consumed, capped by `budget`. type: integer format: int64 facts_distilled: description: Facts extracted from the consumed batches. type: integer format: int64 facts_added: description: Distilled facts that were new. type: integer format: int64 facts_merged: description: Distilled facts folded into a fact already held. type: integer format: int64 facts_dropped_untyped: description: Facts discarded because they matched no declared artifact type. type: integer format: int64 edges_added: description: New relationships recorded between artifacts. type: integer format: int64 artifacts_pruned: description: Artifacts removed as superseded. type: integer format: int64 artifact_files_written: description: How many were written. type: integer format: int64 artifact_files_deleted: description: How many were removed. type: integer format: int64 artifact_count: description: Artifacts the context holds after the run. type: integer format: int64 artifact_bytes: description: Total artifact size after the run. type: integer format: int64 duration_ms: description: Wall time for the groom. type: integer format: int64 required: - status - budget - batches_staged - batches_consumed - facts_distilled - facts_added - facts_merged - facts_dropped_untyped - edges_added - artifacts_pruned - artifact_files_written - artifact_files_deleted - artifact_count - artifact_bytes - duration_ms CurateRunOutput: title: Curate run description: >- What a curate run built. The index it wrote goes live only when `flipped` is true. type: object properties: status: description: How the run ended. type: string flipped: description: >- Whether the new index version was made live. False means queries still read the previous one. type: boolean versioned: description: Whether the run wrote version stamps. type: boolean mode: description: >- Set to `incremental` when the context has a manifest, `full` otherwise. type: string scope: description: What the run covered, e.g. `full` or `reconcile`. type: string images_skipped_no_vision: description: Images left unread because the configured model cannot see. type: integer format: int64 sources_skipped_unreadable: description: Sources no leg could read, skipped rather than failed. type: integer format: int64 usage: $ref: '#/components/schemas/CurateUsage' sample_queries: description: Example questions the corpus can answer type: array items: description: One example question. type: string delta: description: What the run changed, counted per leg. type: object properties: new: description: Sources not seen by a previous curate. type: integer format: int64 changed: description: Sources whose content hash moved. type: integer format: int64 deleted: description: Sources dropped from the index. type: integer format: int64 processed: description: Sources the run actually read. type: integer format: int64 failed: description: Sources that could not be curated. type: integer format: int64 failed_sources: description: Paths of the failed sources, capped in length. type: array items: description: A source path that failed. type: string failed_sources_truncated: description: >- Whether failed_sources was capped and the full list spilled to a file type: boolean failed_sources_file: description: Where the full failed-sources list was written when truncated type: string corpus_artifacts: description: Artifacts written by the corpus-wide reduce pass. type: integer format: int64 corpus_edges: description: Edges written by that pass. type: integer format: int64 corpus_artifacts_screened: description: Corpus artifacts rejected by the quality screen. type: integer format: int64 screened: description: Sum of the per-leg and corpus screen counts type: integer format: int64 chunks_created: description: Chunks written for the first time. type: integer format: int64 chunks_updated: description: Chunks rewritten from changed sources. type: integer format: int64 artifacts_created: description: Per-document artifacts written for the first time. type: integer format: int64 artifacts_updated: description: Per-document artifacts rewritten from changed sources. type: integer format: int64 chunks_screened: description: Chunks rejected by the quality screen. type: integer format: int64 artifacts_screened: description: Per-document artifacts rejected by the quality screen. type: integer format: int64 images_skipped_no_vision: description: Images left unread because the configured model cannot see. type: integer format: int64 sources_skipped_unreadable: description: Sources no leg could read, skipped rather than failed. type: integer format: int64 artifacts_errored: description: Extraction calls that failed outright. type: integer format: int64 sources_content_truncated: description: >- Sources whose text was cut at the input cap, so their tail went unread. type: integer format: int64 required: - new - changed - deleted - processed - failed - failed_sources - failed_sources_truncated - failed_sources_file - corpus_artifacts - corpus_edges - corpus_artifacts_screened - screened - chunks_created - chunks_updated - artifacts_created - artifacts_updated - chunks_screened - artifacts_screened - images_skipped_no_vision - sources_skipped_unreadable - artifacts_errored - sources_content_truncated required: - status - flipped - versioned - mode - scope - images_skipped_no_vision - sources_skipped_unreadable - usage - sample_queries - delta CurateManifestTuneOutput: title: Curate manifest_tune description: >- The post-optimize artifacts re-eval. Targets optimize trial namespaces, so `flipped` is always false and `mode` is always `manifest_tune`. type: object properties: status: description: How the run ended. type: string flipped: description: >- Always false — the run targets optimize trial namespaces, not the live index. x-enum: - false type: boolean versioned: description: Whether the run wrote version stamps. type: boolean mode: description: Always `manifest_tune`. x-enum: - manifest_tune type: string usage: $ref: '#/components/schemas/CurateUsage' delta: description: What the re-eval changed. type: object properties: artifacts_reevaluated: description: >- Artifacts re-extracted under the tuned manifest. Replaces the source counts a normal run reports. type: integer format: int64 failed: description: Sources that could not be re-evaluated. type: integer format: int64 failed_sources: description: Paths of the failed sources, capped in length. type: array items: description: A source path that failed. type: string failed_sources_truncated: description: >- Whether failed_sources was capped and the full list spilled to a file. type: boolean failed_sources_file: description: Where the full failed-sources list was written when truncated. type: string corpus_artifacts: description: Artifacts written by the corpus-wide reduce pass. type: integer format: int64 corpus_edges: description: Edges written by that pass. type: integer format: int64 chunks_created: description: Chunks written for the first time. type: integer format: int64 chunks_updated: description: Chunks rewritten from changed sources. type: integer format: int64 artifacts_created: description: Per-document artifacts written for the first time. type: integer format: int64 artifacts_updated: description: Per-document artifacts rewritten from changed sources. type: integer format: int64 chunks_screened: description: Chunks rejected by the quality screen. type: integer format: int64 artifacts_screened: description: Per-document artifacts rejected by the quality screen. type: integer format: int64 images_skipped_no_vision: description: Images left unread because the configured model cannot see. type: integer format: int64 sources_skipped_unreadable: description: Sources no leg could read, skipped rather than failed. type: integer format: int64 artifacts_errored: description: Extraction calls that failed outright. type: integer format: int64 sources_content_truncated: description: Sources whose text was cut at the input cap. type: integer format: int64 required: - artifacts_reevaluated - failed - failed_sources - failed_sources_truncated - failed_sources_file - corpus_artifacts - corpus_edges - chunks_created - chunks_updated - artifacts_created - artifacts_updated - chunks_screened - artifacts_screened - images_skipped_no_vision - sources_skipped_unreadable - artifacts_errored - sources_content_truncated required: - status - flipped - versioned - mode - usage - delta CurateNoopOutput: title: Curate no-op description: >- Nothing changed since the last run; `delta` is the literal string `empty` rather than a count block, which is what makes this a shape of its own. type: object properties: status: description: How the run ended. type: string flipped: description: Whether a new index version was made live. type: boolean versioned: description: Whether the run wrote version stamps. type: boolean mode: description: >- Set to `incremental` when the context has a manifest, `full` otherwise. type: string scope: description: What the run would have covered. type: string delta: description: Always the literal string `empty`. x-enum: - empty type: string required: - status - flipped - versioned - mode - scope - delta ExploreOutput: title: Explore description: >- What an explore run proposed. Persists nothing — apply it by writing the manifest and forcing a curate. type: object properties: status: description: How the run ended. type: string matches: description: >- Validated template recommendations; unknown or low-confidence proposals are dropped upstream. type: array items: $ref: '#/components/schemas/TemplateMatch' none: description: True when no template fit type: boolean guide: description: The guidance the proposal was steered with, echoed back. type: string models: description: Tier → resolved model id. type: object additionalProperties: description: The model id the tier resolved to. type: string unknown_models: description: >- Model selections dropped as unknown before dispatch, so a caller whose pin was silently ignored can see it. Absent when nothing was dropped. type: array items: description: >- A requested model id or tier name the project's catalog does not carry. type: string usage: $ref: '#/components/schemas/ChatUsage' required: - status - matches - none - guide - models - usage ProfileEstimateOutput: title: Profile estimate description: >- An estimate of what curating the candidate manifest would cost, extrapolated from a sample of the corpus. Persists nothing. type: object properties: state: description: How the run ended. type: string g: description: >- Estimated distinct-artifact count with its uncertainty band. This is what drives the cost estimate, since extraction cost scales with it. type: object properties: value: description: Point estimate of the distinct-artifact count. type: integer format: int64 lo: description: Low end of the band. type: integer format: int64 hi: description: High end of the band. type: integer format: int64 method: description: >- The estimator the recurrence blend selected; `empty` on the no-sample path type: string recurrence: description: >- How often the same subject recurs across sampled documents. Drives which estimator is chosen. type: number format: double confidence: description: Qualitative confidence in the estimate. type: string note: description: Caveat on how to read the estimate. type: string required: - value - lo - hi - method - recurrence - confidence - note density: description: Measured on the sampling window, not the full curate read-depth. type: object properties: entities_per_doc: description: Mean distinct subjects found per sampled document. type: number format: double avg_doc_chars: description: Mean source length in characters. type: integer format: int64 required: - entities_per_doc - avg_doc_chars cost: description: >- Estimated token and time cost of the curate, one estimate per model tier. type: object properties: by_tier: description: One estimate per model tier. type: object properties: lite: $ref: '#/components/schemas/ProfileTierCost' standard: $ref: '#/components/schemas/ProfileTierCost' pro: $ref: '#/components/schemas/ProfileTierCost' required: - lite - standard - pro default_tier: description: The tier the context's current `artifact_model` resolves to type: string required: - by_tier - default_tier sampling: description: >- How much of the corpus the profile actually read. The estimate is extrapolated from this much, so a small `fraction` means a wide band. type: object properties: n_total: description: Sources in the corpus. type: integer format: int64 n_sampled: description: Sources the profile read. type: integer format: int64 fraction: description: The `n_sampled` count over `n_total`. type: number format: double wall_s: description: Seconds the sampling itself took. type: number format: double required: - n_total - n_sampled - fraction - wall_s required: - state - g - density - cost - sampling ProfileEmptyOutput: title: Profile empty description: >- The profile had nothing to sample — a context with no sources, or none the manifest covers. type: object properties: state: description: How the run ended. type: string reason: description: Why no estimate could be made. type: string required: - state - reason ImportOutput: title: Import description: >- What an import staged. Staging only — the sources are not indexed until a curate runs. type: object properties: status: description: How the run ended. type: string imported_items: description: Records pulled from the provider. type: integer format: int64 output_files: description: >- Files written into the source tree. Higher than `imported_items` when archives or multi-page documents expanded. type: integer format: int64 failed: description: Records that could not be ingested. type: integer format: int64 skipped: description: >- Records passed over, e.g. an unsupported extension or a cap already reached. type: integer format: int64 skipped_items: description: Bounded sample of the skipped records type: array items: $ref: '#/components/schemas/ImportManifestEntry' manifest_sample: description: Bounded sample of the records that produced output type: array items: $ref: '#/components/schemas/ImportManifestEntry' manifest_truncated: description: Whether either sample was cut type: boolean manifest_file: description: Where the full manifest was written type: string manifest_object: description: Object-store key of the full manifest type: string required: - status - imported_items - output_files - failed - skipped - skipped_items - manifest_sample - manifest_truncated - manifest_file StatusErrorOutput: title: Status-keyed failure description: >- What a groom or curate reports when it fails terminally: best-effort detail on a task left running so it can be resumed. type: object properties: status: description: How the run ended. type: string error: description: What went wrong. type: string required: - status - error StateFailedOutput: title: State-keyed failure description: The pack / restore terminal-failure convention. type: object properties: state: description: How the run ended. type: string error: description: What went wrong. type: string required: - state - error ErrorOutput: title: API-written failure description: >- What the API itself writes when a runtime reports a failure without an output of its own. Carries no `status` or `state` key. type: object properties: error: description: What went wrong. type: string required: - error ImportLimits: description: The caps the import ran under. type: object properties: allow_archives: description: Whether the run was permitted to expand archives. type: boolean allowed_extensions: description: File extensions the run was allowed to ingest. type: array items: description: One permitted file extension. type: string remaining_files: description: Files left under the context's file cap when the run started. type: integer format: int64 remaining_bytes: description: Bytes left under the context's byte cap when the run started. type: integer format: int64 max_file_bytes: description: Upper bound on one stored object. type: integer format: int64 Manifest: description: >- The document that drives curate and search. Sparse: every field is optional and an omitted one falls back to its default, so absent fields are omitted rather than nulled. A change reaches the index on the next curate; rebuilding what is already indexed needs a forced one. type: object properties: curate: $ref: '#/components/schemas/CurateManifest' optimize: $ref: '#/components/schemas/OptimizeManifest' search: $ref: '#/components/schemas/SearchManifest' CandidateQuery: description: A query the optimize run tunes toward. type: object properties: ask: description: The question to tune retrieval against. type: string answer: description: Used as eval ground truth type: string fallback_to_chunks: description: >- Whether answering this query had to drill from artifacts down to chunks. type: boolean latency_ms: description: Latency the query showed in real traffic, as a tuning baseline. type: number format: double PackDetails: nullable: true description: >- LLM-written corpus summary, echoed from the pack's `details.json`. Null when enrichment was skipped or failed. type: object properties: description: description: Prose summary of what the packed corpus covers. type: string sample_queries: description: Example questions the packed corpus can answer. type: array items: description: One example question. type: string required: - description - sample_queries StageProgress: description: How far one restore phase has got. type: object properties: phase: description: Which phase is running, e.g. `stage_sources`. type: string done: description: Items finished in this phase. type: integer format: int64 total: description: Items this phase will process. type: integer format: int64 required: - phase - done - total ChatUsage: description: >- What a run billed on the chat seam, in tokens. Reported by the inference client, which carries the cache-token fields. Used by optimize and explore. type: object properties: prompt_tokens: description: Input tokens billed across every model. type: integer format: int64 completion_tokens: description: Output tokens billed across every model. type: integer format: int64 total_tokens: description: Input plus output tokens across every model. type: integer format: int64 by_model: description: >- Keyed by the model the proxy actually dispatched to, so tier fallbacks are visible. type: object additionalProperties: $ref: '#/components/schemas/ModelUsage' cache_read_tokens: description: Input served without re-billing. type: integer format: int64 cache_write_tokens: description: Prompt tokens written into cache. type: integer format: int64 required: - prompt_tokens - completion_tokens - total_tokens - by_model - cache_read_tokens - cache_write_tokens CurateUsage: description: >- Curate's token rollup: the chat-seam totals plus the embed/rerank seam (`embed_by_model`). It does not carry the cache-token fields. type: object properties: prompt_tokens: description: Input tokens billed on the chat seam. type: integer format: int64 completion_tokens: description: Output tokens billed on the chat seam. type: integer format: int64 total_tokens: description: Input plus output tokens on the chat seam. type: integer format: int64 by_model: description: >- Chat-seam counts keyed by the model the proxy actually dispatched to. type: object additionalProperties: $ref: '#/components/schemas/ModelUsage' embed_by_model: description: >- Embed and rerank counts keyed by model. These bill separately from the chat totals above and are not included in them. type: object additionalProperties: $ref: '#/components/schemas/SeamUsage' required: - prompt_tokens - completion_tokens - total_tokens - by_model - embed_by_model TemplateMatch: description: One manifest template an explore run proposes for the corpus. type: object properties: template_id: description: Catalog entry the run proposes for this corpus. type: string rationale: description: Why the run thinks this template fits the corpus. type: string confidence: description: >- How sure the run is, 0–1. Low-confidence proposals are dropped before this point. type: number format: double required: - template_id - rationale - confidence ProfileTierCost: description: What the curate would cost at one model tier. type: object properties: tokens: description: >- The `point` estimate sits at G's midpoint and splits into input/output; `lo`/`hi` are band totals. type: object properties: point: description: Estimate at the artifact count's midpoint. type: integer format: int64 lo: description: Total at the low end of the band. type: integer format: int64 hi: description: Total at the high end of the band. type: integer format: int64 input: description: Input share of the point estimate. type: integer format: int64 output: description: Output share of the point estimate. type: integer format: int64 required: - point - lo - hi - input - output time_s: description: Estimated wall-clock band for the curate, in seconds. type: object properties: lo: description: Low end of the wall-clock band. type: integer format: int64 hi: description: High end of the wall-clock band. type: integer format: int64 required: - lo - hi required: - tokens - time_s ImportManifestEntry: description: One ingest record — what happened to a single source the import saw. type: object properties: src: description: The record at the provider. type: string action: description: >- What happened to the record. `parsed` converted the file to text, `copied` took it as-is, `extracted` expanded an archive. x-enum: - parsed - copied - extracted - skipped - failed type: string out: description: Empty exactly when nothing was written type: array items: description: A path written into the source tree. type: string fmt: description: Format the ingester recognized. type: string reason: description: Why it was skipped or how it failed. Empty on success. type: string bytes: description: Bytes written for this record. type: integer format: int64 required: - src - action - out - fmt - reason - bytes CurateManifest: description: >- What curate builds out of the sources — the chunk leg, the artifact leg, or both. type: object properties: chunks: $ref: '#/components/schemas/ChunksConfig' artifacts: $ref: '#/components/schemas/ArtifactsConfig' OptimizeManifest: description: >- The scheduled self-tuning loop: it clusters the queries that answered badly and tries candidate manifests against an ephemeral index until one reproduces the recorded answers. type: object properties: schedule: description: Cron expression the tuning loop runs on. default: 0 * * * * type: string maxLength: 128 latency_threshold_ms: description: >- A query slower than this counts as a problem worth tuning for, as does any query that fell back from artifacts to chunks. `0` disables the latency test, leaving only fallback. default: 60000 type: integer minimum: 0 maximum: 600000 min_group_size: description: >- How many near-duplicate problem queries must cluster together before the loop tunes for them. Keeps a one-off slow query from triggering a manifest change. default: 2 type: integer minimum: 1 maximum: 100 eval_pass_rate_threshold: description: >- Fraction of eval queries a candidate manifest must answer correctly to be considered ready. `1` demands all of them. default: 1 type: number format: double minimum: 0 maximum: 1 max_iterations: description: >- How many candidate manifests the loop tries before stopping with its best. default: 20 type: integer minimum: 1 maximum: 100 SearchManifest: description: >- Standing instructions for the query runtime, pinned on the context rather than sent per turn. type: object properties: instructions: description: >- The context's default system prompt, applied to every turn in scope. A session's own `system_prompt` appends to it rather than replacing it. type: string ModelUsage: description: Per-model chat-seam counts; both usage roll-ups write every key. type: object properties: tier: description: Which tier resolved to this model, so a tier fallback is visible. type: string prompt_tokens: description: Input tokens billed to this model. type: integer format: int64 completion_tokens: description: Output tokens billed to this model. type: integer format: int64 cache_read_tokens: description: Input served without re-billing. type: integer format: int64 cache_write_tokens: description: Prompt tokens written into cache. type: integer format: int64 total_tokens: description: Input plus output tokens for this model. type: integer format: int64 required: - tier - prompt_tokens - completion_tokens - cache_read_tokens - cache_write_tokens - total_tokens SeamUsage: description: >- Per-model embed/rerank-seam counts: billed per token (embed) or per request (rerank). type: object properties: kind: description: >- Which seam this row counts, and therefore which of the two count fields is meaningful. x-enum: - embed - rerank type: string tier: description: Which tier resolved to this model. type: string embed_tokens: description: 'Tokens embedded. Meaningful for `kind: embed`.' type: integer format: int64 rerank_units: description: 'Rerank requests made. Meaningful for `kind: rerank`.' type: integer format: int64 required: - kind - tier - embed_tokens - rerank_units ChunksConfig: description: >- The chunk leg — sources split into passages, embedded for semantic search and optionally indexed for keyword search. type: object properties: enabled: description: Whether curate builds the chunk leg at all. default: true type: boolean embedding_model: description: Model that embeds the chunks. default: multilingual-e5-large type: string maxLength: 128 chunking: $ref: '#/components/schemas/ChunkingConfig' keyword: description: >- The lexical index built alongside the vectors, for exact-term matching. type: object properties: enabled: description: Whether the lexical index is built. default: true type: boolean ArtifactsConfig: description: >- The artifact leg: knowledge an LLM distills out of the sources, as prose files or database rows. Off by default. type: object properties: enabled: description: Whether curate extracts artifacts at all. default: false type: boolean artifact_model: description: >- Model tier that does the extraction. `standard` reads more carefully at a higher token cost. default: lite x-enum: - standard - lite type: string artifact_types: description: >- What kinds of artifact to extract. Nothing is extracted until at least one is declared. type: array items: $ref: '#/components/schemas/ArtifactType' maxItems: 32 edge_types: description: >- Typed, directed relationships between artifact types, which turn the artifacts into a graph the agent can traverse. type: array items: $ref: '#/components/schemas/EdgeType' maxItems: 32 min_doc_count: description: >- How many source documents must mention a corpus-scoped subject before it earns an artifact. Raise it to suppress one-off mentions. An artifact type may override it. default: 1 type: integer minimum: 1 maximum: 64 max_tokens: description: Output cap for one extracted artifact, in tokens. default: 1500 type: integer minimum: 128 maximum: 8192 max_doc_chars: description: >- Input cap for one extraction call, in characters. With windowing off this also caps how much of a source is read at all; everything beyond it is dropped. default: 60000 type: integer minimum: 1000 maximum: 2000000 extraction_window_chars: description: >- Window size for walking a long source across several extraction calls, so its back half is covered rather than dropped. `0` opts out and reads only the head, up to `max_doc_chars`. default: 0 type: integer minimum: 0 maximum: 1000000 mention_max_chars: description: >- Cap on one recorded mention — what a single document says about the subject — in characters. default: 400 type: integer minimum: 100 maximum: 4000 max_mentions_per_artifact: description: >- How many mentions are fed into the pass that reduces them into one corpus-wide artifact. default: 40 type: integer minimum: 1 maximum: 500 mention_context_chars: description: >- Cap on those mentions once concatenated, in characters. Applied after `max_mentions_per_artifact`. default: 8000 type: integer minimum: 1000 maximum: 100000 max_artifacts_per_type: description: How many artifacts one type may produce. Omit for no cap. default: 10000 type: integer minimum: 1 maximum: 10000 ChunkingConfig: description: How a source is cut into chunks. type: object properties: strategy: description: >- Which structure the splitter cuts on — markdown headings, fixed-size windows, HTML elements, CSV rows, or code syntax. Pick the one that matches the corpus. default: markdown_heading x-enum: - markdown_heading - sliding_window - html - csv - code type: string target_size: description: >- Chunk size the splitter aims for, in tokens. Smaller chunks retrieve more precisely; larger ones carry more surrounding context. default: 512 type: integer minimum: 64 maximum: 4096 overlap: description: >- Tokens each chunk repeats from its predecessor, so a passage split across a boundary is still retrievable whole. default: 64 type: integer minimum: 0 maximum: 1024 respect_sections: description: >- Never cut across a section boundary, even when that leaves a chunk short of `target_size`. default: true type: boolean ArtifactType: description: >- One kind of artifact to extract. `name` is required of a written manifest; a read echoes a stored document that lacks one rather than failing. type: object properties: name: description: >- What the type is called. Edge endpoints reference it, and a sqlite type's table is named after it. type: string maxLength: 64 kind: description: >- What the artifact represents, which selects the extraction prompt. `glossary` is the exception — it reads the headings named in `sections` deterministically, with no LLM call. default: topic x-enum: - summary - topic - entity - event - doc - page - glossary type: string scope: description: >- How widely the extraction reaches. `document` extracts one artifact per source; `corpus` reduces mentions from across every source into one artifact per subject. default: corpus x-enum: - document - corpus type: string icon: description: Symbol to display beside artifacts of this kind. type: string maxLength: 64 description: description: >- What this type should capture. Goes into the extraction prompt, so it steers the result. type: string maxLength: 512 coverage: description: >- Topics the extraction is told to address wherever the source speaks to them. type: array items: description: One topic the extraction should address. type: string maxLength: 128 maxItems: 32 sections: description: >- Exact section headings a `glossary`-kind type reads verbatim. Ignored by every other kind. type: array items: description: One exact section heading to read. type: string maxLength: 128 maxItems: 64 min_doc_count: description: Overrides the manifest-level `min_doc_count` for this type. default: 1 type: integer minimum: 1 maximum: 64 format: description: >- Where the artifacts of this type are written — `markdown` a prose file each, `sqlite` rows in the context's structured database. default: markdown x-enum: - markdown - sqlite type: string columns: description: >- Table schema for a `sqlite`-format type. Ignored for `markdown`. Two names that normalize to the same column collide, and the first declaration wins. type: array items: description: One column of the type's table. type: object properties: name: description: Column name. type: string maxLength: 64 type: description: SQLite column affinity. default: TEXT x-enum: - TEXT - INTEGER - REAL - NUMERIC type: string description: description: What the column holds. Steers the extraction that fills it. type: string maxLength: 256 required: - name maxItems: 64 natural_key: description: >- Columns that identify a row uniquely, so re-curating a source upserts its rows rather than duplicating them. `sqlite`-format types only. type: array items: description: One column of the upsert key. type: string maxLength: 64 maxItems: 16 required: - name EdgeType: description: >- One typed, directed relationship between artifact types. `name`, `from`, and `to` are required of a written manifest; a read echoes a stored document that lacks one rather than failing. type: object properties: name: description: What the relationship is called, e.g. `mentions`. type: string maxLength: 64 from: description: >- Source artifact-type name. May equal `to` for a self-referential edge. type: string maxLength: 64 to: description: Target artifact-type name type: string maxLength: 64 description: description: What the relationship means. Steers the extraction that finds it. type: string maxLength: 512 attributes: description: Extra properties recorded on each edge, e.g. `count`. type: array items: description: One attribute name. type: string maxLength: 64 maxItems: 16 required: - name - from - to securitySchemes: bearerAuth: type: http scheme: bearer bearerFormat: JWT description: >- Session token from `POST /auth/login`, sent as `Authorization: Bearer `. apiKey: type: apiKey in: header name: Api-Key description: Pinecone API key, accepted as an alternative to the bearer token. ````