Premium

CustomProviders

Point Plexon at any OpenAI-compatible or Anthropic-compatible endpoint: local LLMs via Ollama, LM Studio, and vLLM; corporate gateways like Azure OpenAI; or alternative backends such as Moonshot's Anthropic-compatible endpoint. Per-model protocol, per-model capability flags, one-click Test Connection, Fetch Models, and JSON import/export.

Download

Included with Premium

Custom Providers (including local LLM support via Ollama, LM Studio, and vLLM) are available on the Premium plan. Pro does not include custom providers.

View Pricing

Bring Any OpenAI- or Anthropic-Compatible Endpoint

Point Plexon at any endpoint that speaks the OpenAI Chat Completions or Anthropic Messages protocol. Each model in the provider declares its own wire format, so you can mix protocols inside a single custom provider: an OpenAI-compatible chat model alongside an Anthropic-compatible reasoning model, all authenticated through the same base URL.

  • Per-model protocol: OpenAI Chat Completions or Anthropic Messages
  • Custom base URL, optional path prefix, and extra headers
  • API-key or bearer-token auth with per-provider secret storage
  • Same streaming, tool-use, and vision paths as first-party providers

Local LLMs: Ollama, LM Studio, vLLM

Keep every token on your machine. Plexon connects to local runtimes over their OpenAI-compatible ports (Ollama on 11434, LM Studio on 1234, vLLM on 8000) and treats them like any other provider. Works with Llama, Qwen, Mistral, DeepSeek, and every other model you can pull locally.

  • Ollama: one-click Clone from Built-in, point at http://localhost:11434
  • LM Studio: OpenAI-compatible server with model fetch and capability flags
  • vLLM: high-throughput self-hosted inference for single-GPU or multi-GPU rigs
  • Zero data leaves the machine: ideal for regulated code and air-gapped work

Corporate & Enterprise Gateways

Route through Azure OpenAI, self-hosted Anthropic-compatible proxies, or an internal AI gateway that enforces allow-lists, audit logging, and spend controls. Custom headers let you add tenant IDs, deployment names, and compliance tokens without forking the app.

  • Azure OpenAI deployments with api-version headers
  • Self-hosted Anthropic-compat proxies (including Moonshot's /anthropic endpoint)
  • Internal corporate gateways with per-tenant auth headers
  • Works alongside the existing provider list: no exclusive lock-in

Per-Model Capability Flags

Every model carries its own capability matrix: streaming, vision, and reasoning (extended thinking). Flags flow through to the UI so vision uploads, thinking budgets, and streaming toggles only appear for models that actually support them.

  • Streaming: enable or disable SSE streaming per model
  • Vision: attach images only to models that accept them
  • Reasoning: surface extended-thinking controls for reasoning-capable models
  • Prevents silent errors from sending unsupported payloads

One-Click Test Connection & Fetch Models

Paste a base URL, hit Test Connection, and Plexon performs a real round-trip against the endpoint, surfacing auth failures, wrong paths, or missing headers before you commit the config. Fetch Models queries the endpoint's /models list and imports them with sensible capability defaults.

  • Test Connection: live probe with clear error messages
  • Fetch Models: auto-populate the model list from the endpoint
  • Clone from Built-in: start from Anthropic, OpenAI, or another preset and edit
  • No guessing: the provider either works end-to-end or tells you why

Import & Export as JSON

Custom provider definitions are plain JSON. Export a provider to share a vetted gateway config with your team, or import a teammate's provider and be productive in seconds. Nothing proprietary: the schema mirrors the on-disk format used by the daemon.

  • Export a single provider or the whole set as JSON
  • Import JSON from a file picker or clipboard
  • Share gateway configs across a team without copying secrets
  • Secrets stay local: API keys are never included in exports

Under the Hood

Per-Model Protocol

Every model declares openai or anthropic as its wire format. One provider can host both, so you can pair an OpenAI-compatible chat model with an Anthropic-compatible reasoning model on the same base URL.

Test Connection

Live probe against the endpoint catches auth failures, missing headers, and wrong paths before you commit the config. No more silent 401s buried inside a chat turn.

Local LLM Ready

Preset shapes for Ollama (11434), LM Studio (1234), and vLLM (8000). Clone from Built-in to start from Anthropic or OpenAI defaults, then swap the base URL and edit.

Capability Flags

Streaming, vision, and reasoning flags are per-model. The UI only surfaces controls (image upload, thinking budget, stream toggle) for models that actually support them.

JSON Import / Export

Provider definitions are plain JSON. Share a vetted gateway config with your team by exporting, then importing on another machine. Secrets stay local: keys are never included.

Fetch Models

Query the endpoint's /models list and auto-populate the model catalog with sensible capability defaults. Edit any entry afterwards to fine-tune flags.

Privacy-First

Custom providers preserve the same zero-knowledge architecture: your code never touches Plexon's servers. With local LLMs, every token stays on your machine.