openapi: 3.1.0 info: title: Memory Intelligence API description: 'The public REST API for **Memory Intelligence** — turn content into Unified Memory Objects (UMOs) your agents can capture, recall, and verify. ## Authentication All endpoints require an API key in the Authorization header: `Authorization: Bearer mi_sk_your_key_here` Get a key from the developer portal (free during beta). ## SDKs Typed Python and JavaScript SDKs are coming. Today, use this REST API directly, or the MCP server (`pip install memoryintelligence-mcp`) to give an MCP client memory in one command. ## Core methods | Endpoint | Purpose | |---|---| | `POST /v1/process` | Capture content → UMO | | `POST /v1/memories/query` | Semantic search / ask | | `GET /v1/memories` | List UMOs | | `GET /v1/memories/{id}/proof` | Provenance verification | | `GET /v1/memories/{id}/explain` | UMO introspection | | `DELETE /v1/memories/{id}` | Delete with receipt | | `POST /v1/batch` | Batch capture | | `POST /v1/upload` | Media file capture | Aliases (same handlers): `POST /v1/memories` → `/v1/process`; `POST /v1/search` → `/v1/memories/query`. ' version: 1.0.0 servers: - url: https://api.memoryintelligence.io description: Production paths: /health: get: tags: - System summary: Health Check description: 'Returns the operational status of the API and its core dependencies. Use this endpoint to confirm the API is reachable and ready to accept requests. Monitor it in your infrastructure health checks and dashboards. `status` values: - `healthy` — all systems operational - `degraded` — API is reachable but one or more services are impaired - `starting` — API is initializing (model pre-warm in progress) `services.database` reflects whether the Postgres connection pool is alive. `services.embedding` reflects whether the embedding model is loaded and ready. A degraded service means captures will still succeed but may fall back to reduced functionality (e.g. search falls back to keyword-only if embedding is not ready).' operationId: health_check_health_get responses: '200': description: Successful Response content: application/json: schema: {} /v1/process: post: tags: - sdk summary: Process Content description: 'Capture raw content and convert it into a Unified Memory Object (UMO). Input: anything that happened — text, events, decisions, observations. Output: a UMO — a discrete, provenance-stamped, semantically-indexed unit of memory. Raw content is never stored; only meaning persists. Returns UMOCompactResponse by default. Use ?detail=full for the verbose MeaningObjectResponse. Add ?include=embedding to append the 384-dim vector. Provenance is automatic: actor_type and device_id are inferred from the API key, SDK headers, User-Agent, and client IP unless explicitly overridden in the request body. Zero-friction attribution.' operationId: process_content_v1_process_post parameters: - name: detail in: query required: false schema: type: string description: '''compact'' (default) or ''full''' default: compact title: Detail description: '''compact'' (default) or ''full''' - name: include in: query required: false schema: anyOf: - type: string - type: 'null' description: Comma-separated extras, e.g. 'embedding' title: Include description: Comma-separated extras, e.g. 'embedding' - name: authorization in: header required: false schema: type: string title: Authorization - name: x-mi-source in: header required: false schema: anyOf: - type: string - type: 'null' title: X-Mi-Source requestBody: required: true content: application/json: schema: $ref: '#/components/schemas/ProcessRequest' responses: '200': description: Successful Response content: application/json: schema: {} '422': description: Validation Error content: application/json: schema: $ref: '#/components/schemas/HTTPValidationError' /v1/memories: get: tags: - MI API summary: mi.list() — Enumerate the caller's memories description: 'List the caller''s Unified Memory Objects with pagination. Returns a compact list without embeddings. Use `limit` and `offset` for pagination. Filter by `source` (e.g. "api", "slack", "gmail"). **Auth:** `Authorization: Bearer mi_sk_...` API key, or a signed-in portal/console JWT (Bearer or `mi_jwt` cookie) — owner read.' operationId: canonical_list_memories_v1_memories_get parameters: - name: limit in: query required: false schema: type: integer default: 20 title: Limit - name: offset in: query required: false schema: type: integer default: 0 title: Offset - name: source in: query required: false schema: anyOf: - type: string - type: 'null' title: Source - name: redact in: query required: false schema: anyOf: - type: boolean - type: 'null' description: Read-time PII redaction override; None=fail-safe redact unless owner-portal read. See finding R11. title: Redact description: Read-time PII redaction override; None=fail-safe redact unless owner-portal read. See finding R11. - name: authorization in: header required: false schema: anyOf: - type: string - type: 'null' title: Authorization - name: x-mi-source in: header required: false schema: anyOf: - type: string - type: 'null' title: X-Mi-Source responses: '200': description: Successful Response content: application/json: schema: {} '422': description: Validation Error content: application/json: schema: $ref: '#/components/schemas/HTTPValidationError' post: tags: - MI API summary: mi.capture() — Turn raw content into a Unified Memory Object description: 'Capture raw content and transform it into a Unified Memory Object (UMO). The 7-stage pipeline (CAPTURE → NORMALIZE → EXTRACT → ENRICH → PARSE → EMBED → VALIDATE) runs synchronously and returns the UMO on completion. Raw content is discarded after processing when `retention_policy` is `meaning_only` (the default). Use `?detail=full` for the verbose echo (including structured claims) — same contract as `/v1/process`. **Auth:** `Authorization: Bearer mi_sk_...`' operationId: canonical_capture_v1_memories_post parameters: - name: detail in: query required: false schema: type: string description: '''compact'' (default) or ''full''' default: compact title: Detail description: '''compact'' (default) or ''full''' - name: include in: query required: false schema: anyOf: - type: string - type: 'null' description: Comma-separated extras, e.g. 'embedding' title: Include description: Comma-separated extras, e.g. 'embedding' - name: authorization in: header required: false schema: type: string title: Authorization - name: x-mi-source in: header required: false schema: anyOf: - type: string - type: 'null' title: X-Mi-Source responses: '200': description: Successful Response content: application/json: schema: {} '422': description: Validation Error content: application/json: schema: $ref: '#/components/schemas/HTTPValidationError' /v1/memories/query: post: tags: - MI API summary: mi.ask() — Retrieve memories by semantic meaning description: 'Semantic search over the caller''s UMO corpus. Embeds the query, runs pgvector cosine similarity, and reranks by keyword overlap and recency. Returns compact UMO summaries by default; use `?detail=full` for the verbose SearchResponse with complete UMO fields, including structured claims (`svo_triples`) — same contract as `/v1/search`. **Auth:** `Authorization: Bearer mi_sk_...` API key, or a signed-in portal/console JWT (Bearer or `mi_jwt` cookie) — owner read.' operationId: canonical_query_v1_memories_query_post parameters: - name: detail in: query required: false schema: type: string description: '''compact'' (default) or ''full''' default: compact title: Detail description: '''compact'' (default) or ''full''' - name: authorization in: header required: false schema: anyOf: - type: string - type: 'null' title: Authorization - name: x-mi-source in: header required: false schema: anyOf: - type: string - type: 'null' title: X-Mi-Source responses: '200': description: Successful Response content: application/json: schema: {} '422': description: Validation Error content: application/json: schema: $ref: '#/components/schemas/HTTPValidationError' /v1/search: post: tags: - sdk summary: Search Memories description: 'Retrieve memories by semantic meaning. Returns ranked UMOs whose meaning is closest to the query. Results are ordered by: semantic similarity (60%) + keyword overlap (15%) + entity overlap (15%) + recency (10%). Use ?detail=full for the verbose SearchResponse with full per-result explanations.' operationId: search_memories_v1_search_post parameters: - name: detail in: query required: false schema: type: string description: '''compact'' (default) or ''full''' default: compact title: Detail description: '''compact'' (default) or ''full''' - name: authorization in: header required: false schema: anyOf: - type: string - type: 'null' title: Authorization - name: x-mi-source in: header required: false schema: anyOf: - type: string - type: 'null' title: X-Mi-Source requestBody: required: true content: application/json: schema: $ref: '#/components/schemas/SearchRequest' responses: '200': description: Successful Response content: application/json: schema: {} '422': description: Validation Error content: application/json: schema: $ref: '#/components/schemas/HTTPValidationError' /v1/memories/{id}: get: tags: - MI API summary: mi.read() — Retrieve a memory by ID description: 'Retrieve a single Unified Memory Object by its UMO ID. Returns the full UMO including summary, entities, topics, sentiment, provenance hashes, quality score, and timestamps. **Tenant isolation:** Identity derived from API key — only returns the memory if it belongs to the caller. **Auth:** `Authorization: Bearer mi_sk_...`' operationId: canonical_read_v1_memories__id__get parameters: - name: id in: path required: true schema: type: string title: Id - name: redact in: query required: false schema: anyOf: - type: boolean - type: 'null' description: Read-time PII redaction override; None=fail-safe redact unless owner-portal read. See finding R11. title: Redact description: Read-time PII redaction override; None=fail-safe redact unless owner-portal read. See finding R11. - name: authorization in: header required: false schema: type: string title: Authorization - name: x-mi-source in: header required: false schema: anyOf: - type: string - type: 'null' title: X-Mi-Source responses: '200': description: Successful Response content: application/json: schema: {} '422': description: Validation Error content: application/json: schema: $ref: '#/components/schemas/HTTPValidationError' delete: tags: - MI API summary: mi.forget() — Delete a memory (soft-delete + grace window) with a deletion receipt. description: 'mi.forget(): Delete a Unified Memory Object by ID. Soft-deletes the UMO: it is immediately excluded from every read (search, list, get) and a deletion receipt is written as cryptographic proof. A background purge worker hard-deletes the content after a short grace window (recovery safety); after that the content is unrecoverable. The receipt is retained as a deletion log entry. **Tenant isolation:** Identity derived from API key — only deletes the memory if it belongs to the caller. Private — this is the ownership guarantee made visible.' operationId: canonical_forget_v1_memories__id__delete parameters: - name: id in: path required: true schema: type: string title: Id - name: authorization in: header required: false schema: type: string title: Authorization - name: x-mi-source in: header required: false schema: anyOf: - type: string - type: 'null' title: X-Mi-Source responses: '200': description: Successful Response content: application/json: schema: {} '422': description: Validation Error content: application/json: schema: $ref: '#/components/schemas/HTTPValidationError' /v1/memories/{id}/proof: get: tags: - MI API summary: mi.verify() — Confirm a memory is real and receipted description: 'Verify cryptographic receipt for a memory. Accepts UMO ID or content hash. **Tenant isolation:** Identity derived from the API key or JWT — only returns proof if the memory belongs to the caller. **Auth:** `Authorization: Bearer mi_sk_...` API key, or a signed-in portal/console JWT (Bearer or `mi_jwt` cookie).' operationId: canonical_verify_v1_memories__id__proof_get parameters: - name: id in: path required: true schema: type: string title: Id - name: authorization in: header required: false schema: anyOf: - type: string - type: 'null' title: Authorization - name: x-mi-source in: header required: false schema: anyOf: - type: string - type: 'null' title: X-Mi-Source responses: '200': description: Successful Response content: application/json: schema: {} '422': description: Validation Error content: application/json: schema: $ref: '#/components/schemas/HTTPValidationError' /v1/memories/{id}/explain: get: tags: - MI API summary: mi.explain() — Understand what a memory contains description: 'Return plain-language explanation of a memory''s contents. **Tenant isolation:** Identity derived from the API key or JWT — only returns explanation if the memory belongs to the caller. **Auth:** `Authorization: Bearer mi_sk_...` API key, or a signed-in portal/console JWT (Bearer or `mi_jwt` cookie) — owner read.' operationId: canonical_explain_v1_memories__id__explain_get parameters: - name: id in: path required: true schema: type: string title: Id - name: level in: query required: false schema: $ref: '#/components/schemas/ExplainLevelRequest' default: full - name: redact in: query required: false schema: anyOf: - type: boolean - type: 'null' description: Read-time PII redaction override; None=fail-safe redact unless owner-portal read. See finding R11. title: Redact description: Read-time PII redaction override; None=fail-safe redact unless owner-portal read. See finding R11. - name: authorization in: header required: false schema: anyOf: - type: string - type: 'null' title: Authorization - name: x-mi-source in: header required: false schema: anyOf: - type: string - type: 'null' title: X-Mi-Source responses: '200': description: Successful Response content: application/json: schema: {} '422': description: Validation Error content: application/json: schema: $ref: '#/components/schemas/HTTPValidationError' /v1/batch: post: tags: - sdk summary: Batch Process Content description: 'Batch capture — process up to 50 items concurrently in a single request. Each item is independent: one failure does not block the rest. Returns per-item results with aggregate success/failure counts, wrapped in the standard APIEnvelope. Provenance is automatic: actor_type and device_id are inferred from the API key, SDK headers, User-Agent, and client IP — same as single capture. Per-item overrides are respected.' operationId: batch_process_content_v1_batch_post parameters: - name: authorization in: header required: false schema: type: string title: Authorization - name: x-mi-source in: header required: false schema: anyOf: - type: string - type: 'null' title: X-Mi-Source requestBody: required: true content: application/json: schema: $ref: '#/components/schemas/BatchProcessRequest' responses: '200': description: Successful Response content: application/json: schema: {} '422': description: Validation Error content: application/json: schema: $ref: '#/components/schemas/HTTPValidationError' /v1/upload: post: tags: - sdk summary: Upload Media description: "Upload a media file and convert it into a UMO.\n\nAccepts multipart/form-data. The\ \ file is routed through the appropriate\nhandler (Whisper for audio/video, OCR for images, pdfplumber\ \ for PDFs),\nand the extracted text flows through the full intelligence pipeline.\n\nSupported:\n\ \ Audio: mp3, wav, m4a, flac, ogg, aac, opus, webm (Whisper transcription)\n Image: png, jpg,\ \ jpeg, gif, bmp, tiff, webp, heic (OCR)\n Document: pdf \ \ (pdfplumber → pypdf)\n Data: csv, tsv, txt, md, json, jsonl, xlsx, docx (tabular/text\ \ extraction)\n Video: mp4, mov, avi, mkv, webm (not yet extracted — planned)\n\ \n**Tenant isolation:** user_ulid is ALWAYS resolved from API key auth context.\nThe form field\ \ is deprecated and ignored for security." operationId: upload_media_v1_upload_post parameters: - name: authorization in: header required: false schema: type: string title: Authorization - name: x-mi-source in: header required: false schema: anyOf: - type: string - type: 'null' title: X-Mi-Source requestBody: required: true content: multipart/form-data: schema: $ref: '#/components/schemas/Body_upload_media_v1_upload_post' responses: '200': description: Successful Response content: application/json: schema: {} '422': description: Validation Error content: application/json: schema: $ref: '#/components/schemas/HTTPValidationError' components: schemas: BatchProcessRequest: properties: items: items: $ref: '#/components/schemas/BatchItemRequest' type: array maxItems: 50 minItems: 1 title: Items type: object required: - items title: BatchProcessRequest description: Batch capture — process up to 50 items in one call. BatchItemRequest: properties: content: anyOf: - type: string - $ref: '#/components/schemas/EncryptedContentRequest' title: Content retention_policy: $ref: '#/components/schemas/RetentionPolicyRequest' default: meaning_only pii_handling: $ref: '#/components/schemas/PIIHandlingRequest' default: extract_and_redact provenance_mode: $ref: '#/components/schemas/ProvenanceModeRequest' default: standard scope: $ref: '#/components/schemas/ScopeRequest' default: user scope_id: anyOf: - type: string - type: 'null' title: Scope Id source: type: string title: Source default: api metadata: additionalProperties: true type: object title: Metadata tags: anyOf: - items: type: string type: array maxItems: 50 - type: 'null' title: Tags description: Caller-supplied tags merged with pipeline-derived tags on the resulting UMO. Same semantics as ProcessRequest.tags. org_ulid: anyOf: - type: string - type: 'null' title: Org Ulid actor_type: $ref: '#/components/schemas/ActorTypeRequest' description: 'Who/what created this memory: human, agent, or system. Server infers from API key and SDK headers if not set.' default: human device_id: anyOf: - type: string maxLength: 64 - type: 'null' title: Device Id description: Persistent device fingerprint. Server infers from caller signals if not set. type: object required: - content title: BatchItemRequest description: Single item within a batch capture request. ProvenanceModeRequest: type: string enum: - standard - authorship - audit title: ProvenanceModeRequest description: Provenance tracking level. RetentionPolicyRequest: type: string enum: - meaning_only - full - summary_only title: RetentionPolicyRequest description: What to retain after processing. PIIHandlingRequest: type: string enum: - detect_only - extract_and_redact - hash - reject title: PIIHandlingRequest description: How to handle PII. ActorTypeRequest: type: string enum: - human - agent - system title: ActorTypeRequest description: 'WHO created this memory — human, agent, or system. This is a provenance-critical field. It distinguishes content authored by a real person from content generated by an AI agent or synthesised by a system pipeline. The value is recorded in the origin lineage step and surfaces in search results, explain, and verify responses. Values: - `human` — a real user authored the content (default) - `agent` — an AI agent (e.g. SpaceAgent, LLM summariser) produced it - `system` — an automated pipeline or integration generated it (no human or agent in the loop)' ScopeRequest: type: string enum: - user - client - project - team - org - all title: ScopeRequest description: Governance scope for memory isolation. EncryptedContentRequest: properties: ciphertext: type: string title: Ciphertext nonce: type: string title: Nonce tag: type: string title: Tag key_id: type: string title: Key Id algorithm: type: string title: Algorithm default: AES-256-GCM type: object required: - ciphertext - nonce - tag - key_id title: EncryptedContentRequest description: 'AES-256-GCM encrypted content payload sent by the Python SDK. The server derives the decryption key from the API key (PBKDF2, fixed salt) — same derivation the SDK used, so no explicit key exchange is needed.' ExplainLevelRequest: type: string enum: - none - human - audit - full title: ExplainLevelRequest description: Explanation level. Body_upload_media_v1_upload_post: properties: file: type: string contentMediaType: application/octet-stream title: File user_ulid: anyOf: - type: string - type: 'null' title: User Ulid org_ulid: anyOf: - type: string - type: 'null' title: Org Ulid scope: anyOf: - type: string - type: 'null' title: Scope default: user actor_type: anyOf: - type: string - type: 'null' title: Actor Type description: 'Override actor type: ''human'', ''agent'', or ''system''. Inferred from headers if omitted.' device_id: anyOf: - type: string - type: 'null' title: Device Id description: Override device ID. Inferred from X-MI-Device-ID header or fingerprinted if omitted. capture_trigger: anyOf: - type: string - type: 'null' title: Capture Trigger description: 'Capture trigger: ''explicit'', ''middleware'', ''decorator'', ''source_sync''.' default: explicit metadata_json: anyOf: - type: string - type: 'null' title: Metadata Json claim_granular: type: boolean title: Claim Granular description: Persist the extracted content (OCR / transcript / text) as a parent UMO plus one child UMO per claim (#446). Default on for uploads — a whiteboard photo, voice note, or meeting recording is the multi-claim case. Structured (spreadsheet) uploads ignore this. default: true claim_level: type: string title: Claim Level description: 'Claim boundary when claim_granular: ''sentence'' | ''compound'' (recommended) | ''assertion''.' default: compound type: object required: - file title: Body_upload_media_v1_upload_post HTTPValidationError: properties: detail: items: $ref: '#/components/schemas/ValidationError' type: array title: Detail type: object title: HTTPValidationError ValidationError: properties: loc: items: anyOf: - type: string - type: integer type: array title: Location msg: type: string title: Message type: type: string title: Error Type input: title: Input ctx: type: object title: Context type: object required: - loc - msg - type title: ValidationError SearchRequest: properties: query: type: string maxLength: 10000 minLength: 1 title: Query scope: $ref: '#/components/schemas/ScopeRequest' default: user scope_id: anyOf: - type: string - type: 'null' title: Scope Id explain: $ref: '#/components/schemas/ExplainLevelRequest' description: 'Explanation level. On the default compact response, any level other than ''none'' also includes the per-component `scores` breakdown on each result (opt-in since #538; default responses carry only the composite `score`).' default: none limit: type: integer maximum: 100.0 minimum: 1.0 title: Limit default: 10 offset: type: integer minimum: 0.0 title: Offset default: 0 date_from: anyOf: - type: string format: date-time - type: 'null' title: Date From date_to: anyOf: - type: string format: date-time - type: 'null' title: Date To tags: anyOf: - items: type: string type: array - type: 'null' title: Tags description: Filter results by tag (any-of overlap). Accepted as `tags` (preferred, matches the capture-side field) or `topics` (legacy alias). Empty list or omitted = no tag filter. entities: anyOf: - items: type: string type: array - type: 'null' title: Entities min_score: anyOf: - type: number maximum: 1.0 minimum: 0.0 - type: 'null' title: Min Score description: 'Optional minimum SEMANTIC similarity (cosine, 0.0-1.0) a result must meet to be returned. Omitted/None = no floor (legacy behavior: always returns top-K, even for out-of-corpus queries). Set this to suppress weak/irrelevant matches — e.g. an auto-recall hook deciding whether any memory is relevant enough to inject. NOTE: tune conservatively; with the current small embedding model, genuinely relevant matches can score modestly, so an aggressive floor may drop real results.' source: anyOf: - type: string - type: 'null' title: Source description: Filter results by capture source (e.g. 'api', 'slack', 'gmail', 'browser'). Only returns memories from this source. budget_tokens: anyOf: - type: integer - type: 'null' title: Budget Tokens org_ulid: anyOf: - type: string - type: 'null' title: Org Ulid redact: anyOf: - type: boolean - type: 'null' title: Redact description: 'Read-time PII redaction override. Omitted/None = fail-safe default (redact hard PII unless the request is a recognised owner-portal read). `false` = return raw values (owner viewing their own data); `true` = force redaction. Agent/MCP surfaces (X-MI-Source: mcp) are always redacted and ignore this flag. Separate from `pii_handling`, which governs capture-time storage. See finding R11.' type: object required: - query title: SearchRequest description: Semantic search over the caller's UMO corpus. ProcessRequest: properties: content: anyOf: - type: string - $ref: '#/components/schemas/EncryptedContentRequest' title: Content description: Raw text content or an AES-256-GCM encrypted payload from the Python SDK. retention_policy: $ref: '#/components/schemas/RetentionPolicyRequest' description: Controls what is persisted. `meaning_only` stores semantic derivatives only — raw content is never saved. default: meaning_only pii_handling: $ref: '#/components/schemas/PIIHandlingRequest' description: How to handle detected PII before storage. default: extract_and_redact provenance_mode: $ref: '#/components/schemas/ProvenanceModeRequest' description: Level of provenance tracking. `audit` mode stores a full hash chain verifiable via mi.verify(). default: standard scope: $ref: '#/components/schemas/ScopeRequest' default: user scope_id: anyOf: - type: string - type: 'null' title: Scope Id description: Scoped resource ID (e.g. project ULID when scope=project). source: type: string title: Source description: Origin of the content (e.g. 'api', 'slack', 'gmail', 'browser'). Free-form string stored in provenance. default: api capture_trigger: $ref: '#/components/schemas/CaptureTriggerRequest' description: How this capture was initiated. Use 'explicit' when calling mi.capture() directly. The SDK sets 'middleware', 'decorator', or 'source_sync' automatically — you do not need to set this manually. default: explicit actor_type: $ref: '#/components/schemas/ActorTypeRequest' description: WHO created this content — 'human' (real user), 'agent' (AI-generated), or 'system' (automated pipeline). Recorded in the provenance origin step and surfaced in search, explain, and verify responses. This is how MI distinguishes real memories from AI-derived or system-synthesised content. default: human device_id: anyOf: - type: string maxLength: 64 - type: 'null' title: Device Id description: Persistent device identifier (e.g. iOS identifierForVendor, a self-issued ULID stored in Keychain). Ties a capture to a specific physical device for provenance. Optional — omit if the source is a server-side integration. metadata: additionalProperties: true type: object title: Metadata description: Arbitrary key-value pairs stored in the UMO. Useful for application-level tagging. tags: anyOf: - items: type: string type: array maxItems: 50 - type: 'null' title: Tags description: Caller-supplied tags merged with pipeline-derived tags on the resulting UMO. Used by /v1/memories/query tag filter. Free-form strings; convention encouraged (e.g. 'signal', 'competitive', ''). Up to 50. org_ulid: anyOf: - type: string - type: 'null' title: Org Ulid description: Organisation ULID for multi-tenant deployments. enable_chunking: type: boolean title: Enable Chunking description: Enable automatic chunking for large content. Creates a parent UMO with child chunks. default: false chunk_threshold: type: integer maximum: 50000.0 minimum: 500.0 title: Chunk Threshold description: Character threshold for chunking. Content exceeding this triggers chunking. default: 3000 claim_granular: type: boolean title: Claim Granular description: Persist long content as a parent UMO plus one child UMO per claim (lineage-linked). Each claim is independently retrievable. Set by mi_capture by default. default: false claim_level: type: string title: Claim Level description: 'Claim boundary granularity when claim_granular=true: ''sentence'' (one atom per sentence), ''compound'' (sentence base, split only genuine multi-assertion sentences — recommended), or ''assertion'' (one atom per clause).' default: compound type: object required: - content title: ProcessRequest description: 'Capture raw content and convert it into a Unified Memory Object (UMO). The 7-stage pipeline (CAPTURE → NORMALIZE → EXTRACT → ENRICH → PARSE → EMBED → VALIDATE) runs synchronously. Raw content is discarded after processing when `retention_policy` is `meaning_only` (the default).' CaptureTriggerRequest: type: string enum: - explicit - middleware - decorator - source_sync title: CaptureTriggerRequest description: 'How this capture was initiated. Stored in UMO metadata and used for: - Search reranking: `explicit` captures rank above ambient ones by default. - Observability: breakdown of capture modes in usage metrics. - Filtering: developers can exclude middleware captures from certain queries. Values: - `explicit` — developer called mi.capture() directly (default) - `middleware` — captured automatically by mi.CaptureMiddleware - `decorator` — captured by the @mi.capture function decorator - `source_sync` — captured by a connected source (Slack, Gmail, etc.)' securitySchemes: HTTPBearer: type: http scheme: bearer