Sign inSign up

teamintelligence/inference:0.1.0-dev.2

Multi-platform
Manifest digest

sha256:e11782a1f8211aa96906e819cf05a99f684e1a1039632398015ca2b27b36c7e7

Last pushed

about 22 hours by ankurrustagi75524

Type

Sandbox Kit

Manifest digest

sha256:e11782a1f8211aa96906e819cf05a99f684e1a1039632398015ca2b27b36c7e7

yaml
schemaVersion: "3"
displayName: Docker Inference
description: Use Docker-hosted models with your coding agent.
kind: mixin
integrates:
    - claude
    - codex
    - opencode
    - hermes
    - hermes-agent
capabilities:
    - type: com.docker.sandbox/credential@1
      config:
        apiKey:
            inject:
                - domain: ${{ kit.args.gateway_host }}
                  format: Bearer %s
                  header: Authorization
            name: DOCKER_INFERENCE_TOKEN
            proxyManaged: true
        phase: runtime
        service: sbx-login
    - type: com.docker.sandbox/network-policy@1
      config:
        runtime:
            allow:
                - ${{ kit.args.gateway_host }}:443
    - type: com.docker.sandbox/lifecycle@1
      config:
        install:
            - command:
                - sh
                - /usr/share/docker-managed/configure.sh
                - install
              description: Select the workload harness and configure Docker Inference
              env:
                - DOCKER_MANAGED_HARNESS
                - DOCKER_MANAGED_GATEWAY_HOST
                - DOCKER_MANAGED_CLAUDE_AUTH_TOKEN
                - DOCKER_MANAGED_CLAUDE_DEFAULT_MODEL
                - DOCKER_MANAGED_CLAUDE_SMALL_MODEL
                - DOCKER_MANAGED_CLAUDE_MODELS
                - DOCKER_MANAGED_CLAUDE_CONTEXT_TOKENS
                - DOCKER_MANAGED_CLAUDE_DISABLE_EXPERIMENTAL_BETAS
                - DOCKER_MANAGED_CODEX_CONFIG
                - DOCKER_MANAGED_CODEX_MODELS
                - DOCKER_MANAGED_OPENCODE_CONFIG
                - DOCKER_MANAGED_HERMES_MODEL
                - DOCKER_MANAGED_HERMES_MODELS
                - DOCKER_MANAGED_HERMES_PYTHON
                - SBX_AGENT
                - XDG_CONFIG_HOME
                - HERMES_HOME
              user: agent
        startup:
            - command:
                - sh
                - /usr/share/docker-managed/configure.sh
                - startup
              description: Refresh Claude models or restore the Hermes managed provider after workload startup
              env:
                - DOCKER_MANAGED_HARNESS
                - DOCKER_MANAGED_GATEWAY_HOST
                - DOCKER_MANAGED_CLAUDE_DEFAULT_MODEL
                - DOCKER_MANAGED_CLAUDE_SMALL_MODEL
                - DOCKER_MANAGED_CLAUDE_MODELS
                - DOCKER_MANAGED_CLAUDE_CONTEXT_TOKENS
                - DOCKER_MANAGED_CLAUDE_REFRESH_MODELS
                - DOCKER_MANAGED_CLAUDE_AUTH_TOKEN
                - DOCKER_MANAGED_CLAUDE_DISABLE_EXPERIMENTAL_BETAS
                - HTTP_PROXY
                - HTTPS_PROXY
                - ALL_PROXY
                - NO_PROXY
                - http_proxy
                - https_proxy
                - all_proxy
                - no_proxy
                - SSL_CERT_FILE
                - CURL_CA_BUNDLE
                - HERMES_HOME
                - DOCKER_MANAGED_HERMES_MODEL
                - DOCKER_MANAGED_HERMES_MODELS
                - DOCKER_MANAGED_HERMES_PYTHON
              user: agent
args:
    claude_auth_token:
        default: proxy-managed
        description: Placeholder replaced by SBX's host-side credential proxy.
        env: DOCKER_MANAGED_CLAUDE_AUTH_TOKEN
    claude_context_tokens:
        default: "1048576"
        description: Maximum context window; startup narrows it to the gateway models' limits.
        env: DOCKER_MANAGED_CLAUDE_CONTEXT_TOKENS
    claude_disable_experimental_betas:
        default: "1"
        description: Keep unqualified Anthropic beta features out of requests.
        env: DOCKER_MANAGED_CLAUDE_DISABLE_EXPERIMENTAL_BETAS
    claude_model:
        default: deepseek-ai/deepseek-v4.1-flash-1017
        description: Initial managed model; a saved /model choice takes precedence.
        env: DOCKER_MANAGED_CLAUDE_DEFAULT_MODEL
    claude_models:
        default: |
            [
              { "model": "deepseek-ai/deepseek-v4.1-flash-1017", "label": "DeepSeek V4.1 Flash" },
              { "model": "moonshotai/kimi-k3.1-pro", "label": "Kimi K3.1 Pro" },
              { "model": "zai-org/glm-5.3", "label": "GLM 5.3" },
              { "model": "zai-org/glm-5.3-flash", "label": "GLM 5.3 Flash" }
            ]
        description: Bundled fallback choices and labels; startup refreshes the picker from the gateway.
        env: DOCKER_MANAGED_CLAUDE_MODELS
    claude_refresh_models:
        default: "1"
        description: Refresh model choices from the gateway on each sandbox boot.
        env: DOCKER_MANAGED_CLAUDE_REFRESH_MODELS
    claude_small_model:
        default: zai-org/glm-5.3-flash
        description: Preferred managed model for Haiku/background tasks.
        env: DOCKER_MANAGED_CLAUDE_SMALL_MODEL
    codex_config:
        default: |
            model = "deepseek-ai/deepseek-v4.1-flash-1017"
            model_provider = "docker-managed"
            model_reasoning_effort = "high"
            model_catalog_json = "/home/agent/.codex/docker-managed-models.json"
            web_search = "disabled"

            [model_providers.docker-managed]
            name = "Docker Inference"
            base_url = "https://${DOCKER_MANAGED_GATEWAY_HOST}/v1"
            env_key = "DOCKER_INFERENCE_TOKEN"
            wire_api = "responses"
            requires_openai_auth = false
            supports_websockets = false
        description: Codex provider configuration for the managed gateway.
        env: DOCKER_MANAGED_CODEX_CONFIG
    codex_models:
        default: |
            [
              {
                "id": "deepseek-ai/deepseek-v4.1-flash-1017",
                "name": "DeepSeek V4.1 Flash",
                "input": ["text", "image"],
                "context": 1048576,
                "reasoning": ["low", "medium", "high", "max"]
              },
              {
                "id": "moonshotai/kimi-k3.1-pro",
                "name": "Kimi K3.1 Pro",
                "input": ["text", "image"],
                "context": 1048576,
                "reasoning": []
              },
              {
                "id": "zai-org/glm-5.3",
                "name": "GLM 5.3",
                "input": ["text"],
                "context": 1048576,
                "reasoning": ["none", "minimal", "low", "medium", "high", "xhigh", "max"]
              },
              {
                "id": "zai-org/glm-5.3-flash",
                "name": "GLM 5.3 Flash",
                "input": ["text", "image"],
                "context": 1048576,
                "reasoning": ["none", "minimal", "low", "medium", "high", "xhigh", "max"]
              }
            ]
        description: Managed models and their reviewed Responses API reasoning levels for Codex's picker.
        env: DOCKER_MANAGED_CODEX_MODELS
    gateway_host:
        default: inference.docker.com
        description: Gateway hostname at docker.com or a subdomain; HTTPS on port 443.
        pattern: (?i)^([a-z0-9]([a-z0-9-]{0,61}[a-z0-9])?\.)*docker\.com$
        env: DOCKER_MANAGED_GATEWAY_HOST
    harness:
        default: auto
        description: Harness to configure; auto uses SBX_AGENT or the only installed supported agent.
        enum:
            - auto
            - claude
            - codex
            - opencode
            - hermes
        env: DOCKER_MANAGED_HARNESS
    hermes_model:
        default: deepseek-ai/deepseek-v4.1-flash-1017
        description: Initial Hermes managed model; a saved managed selection takes precedence.
        env: DOCKER_MANAGED_HERMES_MODEL
    hermes_models:
        default: |
            {
              "deepseek-ai/deepseek-v4.1-flash-1017": {
                "context_length": 1048576
              },
              "moonshotai/kimi-k3.1-pro": {
                "context_length": 1048576
              },
              "zai-org/glm-5.3": {
                "context_length": 1048576
              },
              "zai-org/glm-5.3-flash": {
                "context_length": 1048576
              }
            }
        description: Bundled Hermes model IDs and context lengths.
        env: DOCKER_MANAGED_HERMES_MODELS
    hermes_python:
        default: auto
        description: Python with the Hermes YAML library; auto uses the Docker Hermes virtualenv or python3.
        env: DOCKER_MANAGED_HERMES_PYTHON
    opencode_config:
        default: |
            {
              "$schema": "https://opencode.ai/config.json",
              "provider": {
                "docker-managed": {
                  "npm": "@ai-sdk/openai-compatible",
                  "name": "Docker Inference",
                  "options": {
                    "baseURL": "https://{env:DOCKER_MANAGED_GATEWAY_HOST}/v1",
                    "apiKey": "proxy-managed"
                  },
                  "models": {
                    "deepseek-ai/deepseek-v4.1-flash-1017": {
                      "name": "DeepSeek V4.1 Flash",
                      "limit": { "context": 1048576, "output": 262144 },
                      "modalities": { "input": ["text", "image"], "output": ["text"] },
                      "attachment": true
                    },
                    "moonshotai/kimi-k3.1-pro": {
                      "name": "Kimi K3.1 Pro",
                      "limit": { "context": 1048576, "output": 262144 },
                      "modalities": { "input": ["text", "image"], "output": ["text"] },
                      "attachment": true
                    },
                    "zai-org/glm-5.3": {
                      "name": "GLM 5.3",
                      "limit": { "context": 1048576, "output": 262144 }
                    },
                    "zai-org/glm-5.3-flash": {
                      "name": "GLM 5.3 Flash",
                      "limit": { "context": 1048576, "output": 262144 },
                      "modalities": { "input": ["text", "image"], "output": ["text"] },
                      "attachment": true
                    }
                  }
                }
              },
              "model": "docker-managed/deepseek-ai/deepseek-v4.1-flash-1017",
              "small_model": "docker-managed/zai-org/glm-5.3-flash"
            }
        description: OpenCode provider configuration for the managed gateway.
        env: DOCKER_MANAGED_OPENCODE_CONFIG
build: |
    FROM scratch
    COPY configure.sh /usr/share/docker-managed/configure.sh
    COPY adapters/ /usr/share/docker-managed/adapters/