EvolutionaryScale Encoding API

openapi: 3.1.0
info:
  title: EvolutionaryScale Forge ESM3 Embeddings Encoding API
  description: 'Hosted inference API for the ESM3 multimodal protein language model from EvolutionaryScale. ESM3 reasons jointly across sequence, structure, and function tracks. The Forge API exposes generate, batch_generate, encode, decode, forward_and_sample, and logits operations across small (1.4B), medium (7B), and large (98B) parameter checkpoints. All requests authenticate with a Forge API token obtained from forge.evolutionaryscale.ai. This OpenAPI is reconstructed from the open-source `esm` Python SDK (Biohub/esm) which is the canonical client for the Forge service.

    '
  version: 2024-08-01
  contact:
    name: EvolutionaryScale Forge
    url: https://forge.evolutionaryscale.ai
  license:
    name: Cambrian Inference Clickthrough License Agreement
    url: https://www.evolutionaryscale.ai/policies/cambrian-inference-clickthrough-license-agreement
servers:
- url: https://forge.evolutionaryscale.ai
  description: EvolutionaryScale Forge Production
security:
- BearerAuth: []
tags:
- name: Encoding
  description: Tokenize ESMProtein objects into ESMProteinTensor inputs (and back).
paths:
  /api/v1/encode:
    post:
      summary: Encode Protein To Tensor
      description: 'Convert an `ESMProtein` into an `ESMProteinTensor` representation by running the structure tokenizer (and other per-track tokenizers) over the input. The tensor representation is required for low-level forward_and_sample and logits operations.

        '
      operationId: encode
      tags:
      - Encoding
      requestBody:
        required: true
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/EncodeRequest'
      responses:
        '200':
          description: Tokenized protein tensor.
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/ESMProteinTensor'
        4XX:
          $ref: '#/components/responses/ErrorResponse'
  /api/v1/decode:
    post:
      summary: Decode Protein Tensor
      description: 'Inverse of encode. Runs the structure token decoder to convert an `ESMProteinTensor` back into an `ESMProtein` with sequence and atom coordinates.

        '
      operationId: decode
      tags:
      - Encoding
      requestBody:
        required: true
        content:
          application/json:
            schema:
              $ref: '#/components/schemas/DecodeRequest'
      responses:
        '200':
          description: Decoded protein.
          content:
            application/json:
              schema:
                $ref: '#/components/schemas/ESMProtein'
        4XX:
          $ref: '#/components/responses/ErrorResponse'
components:
  schemas:
    DecodeRequest:
      type: object
      required:
      - model
      - protein_tensor
      properties:
        model:
          type: string
          example: esm3-medium-2024-08
        protein_tensor:
          $ref: '#/components/schemas/ESMProteinTensor'
    Error:
      type: object
      properties:
        error:
          type: object
          properties:
            type:
              type: string
            message:
              type: string
            code:
              type: string
    EncodeRequest:
      type: object
      required:
      - model
      - protein
      properties:
        model:
          type: string
          example: esm3-medium-2024-08
        protein:
          $ref: '#/components/schemas/ESMProtein'
    ESMProtein:
      type: object
      description: Raw protein representation with optional sequence, structure, function, secondary structure, and SASA tracks.
      properties:
        sequence:
          type: string
          description: Amino acid sequence using one-letter codes (A, R, N, ..., V). Underscore `_` marks masked positions.
          example: MKTAYIAKQRQISFVK_____SSERVKKLLVGDIVT
        coordinates:
          type: array
          description: Per-residue atom37 coordinates (list of [37,3] arrays).
          items:
            type: array
            items:
              type: array
              items:
                type: number
        secondary_structure:
          type: string
          description: DSSP-style per-residue secondary structure annotation.
        sasa:
          type: array
          description: Per-residue solvent accessible surface area.
          items:
            type: number
        function_annotations:
          type: array
          description: Optional per-residue function annotations.
          items:
            type: object
        ptm:
          type: number
          description: Predicted TM-score (if available).
        plddt:
          type: array
          description: Per-residue pLDDT confidence (if available).
          items:
            type: number
    ESMProteinTensor:
      type: object
      description: Tokenized representation of a protein across all ESM3 tracks.
      properties:
        sequence:
          type: array
          items:
            type: integer
        structure:
          type: array
          items:
            type: integer
        secondary_structure:
          type: array
          items:
            type: integer
        sasa:
          type: array
          items:
            type: integer
        function:
          type: array
          items:
            type: integer
        residue_annotations:
          type: array
          items:
            type: integer
  responses:
    ErrorResponse:
      description: Error returned by the Forge API.
      content:
        application/json:
          schema:
            $ref: '#/components/schemas/Error'
  securitySchemes:
    BearerAuth:
      type: http
      scheme: bearer
      description: Forge API token issued via forge.evolutionaryscale.ai.