emem Find Similar API

openapi: 3.2.0
info:
  description: emem is shared memory for AI agents working together in the real world.
  license:
    name: Apache-2.0
  title: emem Find Similar API
  version: 2.4.0
  x-emem-surface-asymmetry:
    memory_notes: MCP only
    reach_them_at: POST /mcp, method tools/call
    read_side_is_here:
    - /v1/memory/search
    - /v1/memory/sse
    - /memories/{path}
    tools:
    - emem_memory_create
    - emem_memory_view
    - emem_memory_delete
    - emem_memory_rename
    - emem_memory_str_replace
    - emem_memory_supersede
    why_not_here: These write the agent correspondence plane, which is prose and untrusted-by-declaration. It is deliberately not part of the REST fact surface, and the two planes are kept apart rather than merged for convenience.
servers:
- description: Hosted instance (HTTPS-only)
  url: https://emem.dev
tags:
- name: Find Similar
paths:
  /v1/find_similar:
    post:
      description: 'k-NN over the corpus by cell embedding or inline vector. Returns `neighbours` ordered nearest-first, each with `cell64`, `score` and the `band` scanned, plus a signed receipt over the vectors read. Scoring is `mode`: cosine is exact fp32; hamming is a sign-bit popcount that scans far more cells for the same budget; hamming_then_rerank does both. `k` is 1..1000, default 10. It ranks what the corpus already holds; only when the KEY''s own vector is missing does it materialise that one band for the key, signed and reported in `materialize_notes`, then retry. Neighbours are never materialised, so an empty result means nobody has attested a vector nearby, not that nowhere resembles the key.


        When to use: Call when the user asks ''find places like X'', ''where else looks like this'', or hands an embedding to find neighbours. `key` is either a cell64 or `inline:[x,y,...]`. Default band is `geotessera` (128-D Tessera foundation embedding); pass `band: "geotessera.multi_year"` for the 1152-D 9-vintage (2017–2025) fusion.'
      operationId: emem_find_similar
      requestBody:
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/FindSimilarReq'
        required: true
      responses:
        '200':
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/FindSimilarResp'
          description: ok
        default:
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/ErrorEnvelope'
          description: 'error, the emem.error.v1 envelope. Branch on the stable `code` (see GET /v1/errors), not the message. A malformed or missing-field request body returns `code: invalid_argument` with the offending field named in `message`.'
      summary: k-NN over band vectors
      tags:
      - Find Similar
components:
  schemas:
    Cost:
      description: 'Self-declared cost block on every receipt. Honest accounting: latencies are observed, freshness is the age of the stalest source cited (null when undatable, never 0 as a stand-in), `was_cached` is true when the hot cache served the read.'
      properties:
        credits:
          description: Conceptual cost units; 0 for L0/L1 read endpoints on the hosted responder.
          type: number
        latency_p50_ms:
          type: number
        latency_p99_ms:
          type: number
        source_freshness_s:
          description: 'Age of the STALEST source this response cites: now minus the earliest captured_at across the returned facts'' sources. null when nothing in the response carries a dated source, which is the honest answer for a primitive that reads no observation. Was a hardcoded 0 until 2026-08-05, so a 2021 DEM tile reported as 0 s old; a null here means unknown, never fresh.'
          type:
          - integer
          - 'null'
        was_cached:
          type: boolean
      type: object
    ErrorEnvelope:
      description: The `emem.error.v1` failure envelope returned by every endpoint on a 4xx/5xx. Branch on the stable `code` (not the human `message`). See GET /v1/errors for the full code catalog.
      properties:
        code:
          description: Stable machine-readable error code. One of the codes in GET /v1/errors.
          example: invalid_argument
          type: string
        details:
          description: Optional structured recovery hints; present on errors that ship machine-readable next-steps.
          type: object
        message:
          description: Human-readable detail. For invalid_argument this names the offending field (e.g. "missing field `q`").
          type: string
        path:
          description: Request path that produced the error.
          example: /v1/ask
          type: string
        schema:
          const: emem.error.v1
          type: string
      required:
      - code
      - message
      - schema
      type: object
    PubKey:
      description: Ed25519 32-byte public key, base32-nopad-lowercase encoded (52 chars). Returned in receipts and `/.well-known/emem.json`.
      example: 777er3yihgifqmv5hmc2wwmyszgddzderzhsx6rex4yoakwomvka
      type: string
    Cell64:
      description: 'cell64 wire form: four base-65,536 bigrams separated by dots, e.g. `defi.zb4d9.pefa.zf619`. Encoded resolution is ~9.55 m at the equator. Each bigram is either a CVCV quad, consonant `[bcdfghjklmnpqrstvwxyz]` followed by vowel `[aeiouAEIOU]` repeated twice, OR a synthetic 5-char `z[0-9a-f]{4}` slot used for the unused pad cells in the 65,536-entry alphabet. The regex pin matches `pattern` below byte-for-byte and is also surfaced under `Cell64Pattern` so agents can validate before sending.'
      example: defi.zb4d9.pefa.zf619
      maxLength: 23
      minLength: 19
      pattern: ^(?:(?:[bcdfghjklmnpqrstvwxyz][aeiouAEIOU]){2}|z[0-9a-f]{4})(?:\.(?:(?:[bcdfghjklmnpqrstvwxyz][aeiouAEIOU]){2}|z[0-9a-f]{4})){3}$
      type: string
    FindSimilarReq:
      properties:
        band:
          default: geotessera
          description: Vector band to scan. Default geotessera (128-D, int8+scale upstream → decoded f32 over the wire). Pass `geotessera.bin128` (or any band's `.bin128` sibling, plus `mode:"hamming"`) for the binary fast path.
          type: string
        k:
          default: 10
          maximum: 1000
          minimum: 1
          type: integer
        key:
          description: cell64 (look up that cell's vector) or 'inline:[x,y,...]' literal vector
          type: string
        mode:
          default: cosine
          description: Scoring mode. `cosine` (default) is fp32 over the full vector. `hamming` is sign-bit popcount over the binary sibling band, ~1000× faster scan, ~65% recall@10 alone. `hamming_then_rerank` triages with Hamming then re-ranks the top 4·k by cosine, matches cosine precision at ~16× less work.
          enum:
          - cosine
          - hamming
          - hamming_then_rerank
          type: string
      required:
      - key
      type: object
    FactCid:
      description: 'Content id of a fact: base32-nopad-lowercase encoding of `blake3(canonical_cbor(fact))`, the FULL 32-byte digest with no truncation. Always 52 characters, alphabet `[a-z2-7]`. A cid of any other length is a damaged citation, not a shorter address: /v1/memory_token/resolve rejects it as `fact_cid_malformed_length` rather than guessing. Note that `entity_cid` and `bundle_cid` are NOT this shape; both truncate to 16 bytes (26 characters) and hash an identity anchor or a citation list rather than a complete body.'
      example: qtv2bco56qw4pmlohk56dotoxyl3atmnjpmzrijj2kazw2mj57oq
      maxLength: 52
      minLength: 52
      pattern: ^[a-z2-7]{52}$
      type: string
    Receipt:
      description: 'Ed25519-signed receipt. The browser-side verifier at /verify reconstructs the preimage from the receipt fields alone, no callback to the issuer. **A receipt is byte-for-byte or nothing.** Current receipts carry `preimage_version: 2`, whose preimage binds request_id, served_at, primitive, cells, fact_cids AND, when present, the scope / as_of / edges / source_versions / field digests and the `merkle_proof` segment. Reshaping a receipt — dropping a field an SDK considers redundant, re-keying it, summarising it, round-tripping it through a lossy model — invalidates the signature BY DESIGN, and the result is indistinguishable on the wire from tampering. Store and forward the responder''s exact bytes. POST /v1/verify_receipt names which of the two it is where it can prove the difference (`reason: receipt_reshaped_after_signing` with a `failure_detail`). What is NOT signed: the caller''s `place`/`q` string, raw `lat`/`lng`, requested `bands[]`, requested `tslot`, and `intent` — a wrong-place geocode produces a valid signature for the wrong cell. Branch on /v1/locate `selected.is_high_confidence` before trusting place-anchored answers. Also: `fact_cid` is per-replica (signed_at differs across responders even for byte-identical upstream pixels); cross-replica join key is the tuple (cell, band, tslot). /v1/recall_polygon emits one independently signed receipt per cell under `by_cell.<cell>.receipt`, `merged_facts[]` is convenience flattening and is NOT covered by an aggregate signature.'
      properties:
        cells:
          items:
            $ref: '#/components/schemas/Cell64'
          type: array
        cost:
          $ref: '#/components/schemas/Cost'
        fact_cids:
          items:
            $ref: '#/components/schemas/FactCid'
          type: array
        intent:
          description: Optional natural-language hint. Populated when served via /v1/intent.
          type: string
        merkle_proof:
          description: 'Inclusion proof for `fact_cids[0]` when persisted. Omitted from JSON when the cited facts pre-date the proof tree; under preimage_version 2 that absence is itself signed (an explicit ABSENT marker), so it is a statement rather than a gap. Do not strip this field: v2 binds it into the signature and removing it makes an authentic receipt report `signature_valid: false`.'
          properties:
            leaf_index:
              description: u32 leaf index in the canonical-sorted batch.
              type: integer
            path:
              description: Sibling hashes leaf→root.
              items:
                description: 32-byte sibling hash as a byte array
                items:
                  type: integer
                type: array
              type: array
            root:
              description: The expected 32-byte batch root as a byte array.
              items:
                type: integer
              type: array
            version:
              description: 'Merkle hashing rule: 0 (omitted) = legacy unprefixed, 1 = RFC 6962-style prefixed.'
              type: integer
          required:
          - leaf_index
          - path
          - root
          type: object
        primitive:
          description: 'Namespaced wire form: `emem.recall`, `emem.find_similar`, `emem.verify`, …'
          type: string
        registry_cid:
          description: CID of the function registry version in force.
          type: string
        request_id:
          description: ULID generated per request.
          type: string
        responder:
          $ref: '#/components/schemas/PubKey'
        responder_key_epoch:
          description: u32 rotation counter; bumps when the operator rotates keys.
          type: integer
        responder_pubkey_b32:
          $ref: '#/components/schemas/PubKey'
        schema_cid:
          description: CID of the active CDDL profile.
          type: string
        served_at:
          description: ISO 8601 UTC, second precision.
          type: string
        signature:
          description: Ed25519 signature, 64 bytes base32-nopad-lowercase encoded.
          type: string
        source_versions:
          additionalProperties:
            type: string
          description: Per-source freshness map.
          type: object
      required:
      - request_id
      - served_at
      - primitive
      - cells
      - fact_cids
      - schema_cid
      - responder
      - responder_key_epoch
      - responder_pubkey_b32
      - signature
      - registry_cid
      type: object
    FindSimilarResp:
      description: Response of /v1/find_similar. `neighbors` is the top-k list ordered by similarity (descending). `mode` echoes the scoring choice (`cosine` / `hamming` / `hamming_then_rerank`).
      properties:
        band:
          type: string
        mode:
          enum:
          - cosine
          - hamming
          - hamming_then_rerank
          type: string
        neighbors:
          items:
            description: 'Stable neighbor schema: cell/score/lat/lng/place_label_cached are always present. lat/lng are explicit null for inline-vector queries or undecodable cells (no honest centroid), never absent, never fabricated.'
            properties:
              cell:
                $ref: '#/components/schemas/Cell64'
              fact_cid:
                $ref: '#/components/schemas/FactCid'
              label:
                description: Reader-friendly place label, if the cell is named in the gazetteer.
                type: string
              lat:
                description: Centroid latitude decoded from `cell`; null when the cell has no honest centroid (inline vector / undecodable).
                type:
                - number
                - 'null'
              lng:
                description: Centroid longitude decoded from `cell`; null when unknown (see `lat`).
                type:
                - number
                - 'null'
              place_label_cached:
                description: Best-effort gazetteer label (~25 km gate); null when the cell isn't near a known anchor.
                type:
                - string
                - 'null'
              score:
                description: Cosine similarity in [-1, 1] for `cosine` / `hamming_then_rerank`; normalised Hamming agreement in [0, 1] for `hamming`.
                type: number
            required:
            - cell
            - score
            - lat
            - lng
            type: object
          type: array
        receipt:
          $ref: '#/components/schemas/Receipt'
      required:
      - neighbors
      - receipt
      type: object