pi-venice

Pi Coding Agent extension for using Venice.AI as provider. All text, image, edit, video models and tools with support for S3 output storage.

A Pi extension that adds:

• Venice text models to Pi's /model picker
• Venice image/video models & media tools for image generation, image editing, upscale, background removal, and video generation
• configurable local or S3-compatible image/edit/video output storage

Quick start

export VENICE_API_KEY="your-venice-api-key"
pi -e npm:pi-venice

# or from Github

pi -e https://github.com/tunnckoCore/pi-venice

# or direct install

pi install npm:pi-venice

What it supports

Text models

The extension fetches the Venice model catalog and registers Venice text models as a Pi provider named venice.

That means you can use Venice text models through Pi's normal model selection flow, eg. with /model

Included tools

• venice_list_models
• venice_image_generate
• venice_image_edit
• venice_image_multi_edit
• venice_image_upscale
• venice_background_remove
• venice_video_generate
• venice_video_retrieve
• venice_video_complete

Catalog families

The extension can track broader Venice catalog families such as:

• ✅ text
• ✅ image
• ✅ edit
• ✅ upscale
• ✅ video
• ❌ embedding
• ❌ music
• ❌ tts
• ❌ asr
• ❌ audio

NOTE: Not every catalog family has a dedicated Pi tool. Families without runtime support are still visible in catalog/config output so the extension stays honest about what Venice exposes.

Authentication

The extension currently supports API key authentication.

It looks for a Venice API key in either:

• env variables like VENICE_API_KEY (or custom one)
• global pi config auth (~/.config/pi/agent/auth.json or ~/.pi/agent/auth.json)

Example auth.json using an environment variable name:

{
  "venice": {
    "type": "api_key",
    "key": "VENICE_API_KEY"
  }
}

Example auth.json using a literal API key:

{
  "venice": {
    "type": "api_key",
    "key": "vnc_live_abc123"
  }
}

OAuth is not implemented in this extension today. If Venice exposes a stable OAuth flow in the future, it could be added through Pi's custom-provider OAuth hooks and /login, but that is not wired up yet.

Make Venice your default provider

Example ~/.pi/agent/settings.json:

{
  "defaultProvider": "venice",
  "defaultModel": "zai-org-glm-5",
  "enabledModels": ["venice/*"]
}

pi-venice settings

Extension-specific settings live under:

{
  "pi-venice": {
    "...": "..."
  }
}

Supported locations:

• global: ~/.pi/agent/settings.json
• project: .pi/settings.json

Project settings override global settings.

Local output example

{
  "pi-venice": {
    "apiKeyEnv": "VENICE_API_KEY",
    "families": {
      "enabled": ["text", "image", "edit", "video"],
      "defaults": {
        "text": "zai-org-glm-5",
        "image": "flux-2-max",
        "edit": "qwen-edit",
        "video": "seedance-2-0-text-to-video"
      }
    },
    "output": {
      "rootDir": ".pi/venice-output"
    },
    "storage": {
      "files": {
        "adapter": "local",
        "local": {
          "baseDir": ".pi/venice-output"
        }
      }
    }
  }
}

S3 / R2 output example

Output generated images or videos to external S3-compatible storage:

{
  "pi-venice": {
    "storage": {
      "files": {
        "adapter": "s3",
        "s3": {
          "endpoint": "https://<accountid>.r2.cloudflarestorage.com",
          "bucket": "pi-venice-artifacts",
          "region": "auto",
          "prefix": "my-project",
          "forcePathStyle": true,
          "publicBaseUrl": "https://cdn.example.com/pi-venice",
          "credentials": {
            "accessKeyId": "env:R2_ACCESS_KEY_ID",
            "secretAccessKey": "env:R2_SECRET_ACCESS_KEY"
          }
        }
      }
    }
  }
}

Adapter credential references support:

• env:NAME
• literal values - eg. you access key and secret keys inside your Pi config.

Commands

/venice-refresh-models

Refresh the Venice model catalog and re-register Venice text models.

/venice-status

Show:

• enabled catalog families
• implemented provider families
• implemented tool families
• enabled but not actionable families
• file storage adapter
• defaults
• model counts
• last refresh status
• active video jobs

/venice-models [family|all] [limit]

Examples:

/venice-models
/venice-models text 30
/venice-models embedding 20
/venice-models video 20

/venice-defaults

Show or set default model IDs used by the extension.

/venice-families

Show or set enabled Venice catalog families.

Examples:

/venice-families
/venice-families text,image,embedding,video
/venice-families tts,asr,audio
/venice-families all

Tool summary

venice_list_models

List Venice models from the cached catalog.

venice_image_generate

Generate images with /image/generate.

venice_image_edit

Edit a single image with /image/edit.

venice_image_multi_edit

Edit or composite up to 3 images with /image/multi-edit.

venice_image_upscale

Upscale or enhance an image with /image/upscale.

venice_background_remove

Remove an image background with /image/background-remove.

venice_video_generate

Queue a video job, optionally quote it, poll for completion, and save the result.

venice_video_retrieve

Retrieve or continue polling a previously queued video job.

venice_video_complete

Delete a completed remote Venice video job from Venice storage.

Non-interactive support

The extension works in headless and non-TUI flows for core functionality, including:

• provider registration
• model refresh
• media tools
• local file output
• S3-compatible output upload
• session state restoration

UI-only features are guarded so the extension can still be used in print, JSON, and embedded agent workflows.

Output behavior

By default, outputs are written locally under .pi/venice-output.

If pi-venice.storage.files.adapter is set to s3, generated outputs are uploaded to the configured S3-compatible backend instead.

License

Apache-2.0