DL Prism

DL Prism is an AI gateway that runs on your own machine. It gives every AI tool you use one address, http://127.0.0.1:8130/v1, and decides which model answers each request: a cloud model such as Claude or GPT, or a local model running in Ollama, LM Studio or vLLM.

Beta

DL Prism is in beta. The gateway is version 0.2.0 and the desktop companion is version 1.0.0, both from the first tagged release. Settings and endpoints may still change.

What it does

One endpoint for every model. Prism speaks the OpenAI chat API, so any tool that lets you change its base URL can use it: Codex CLI, Cursor, Continue, Windsurf, scripts and SDKs. It also accepts the Anthropic Messages API, so Claude Code can point at it too.
Manual or automatic routing. Ask for a model by name, or ask for auto and let your rules choose. Rules match on the kind of task, the programming language, images, request size, offline mode and budget.
Failover. If a provider is down, rate-limited or returns a server error, Prism tries the next model in the chain you set.
Cost tracking and a monthly budget. Prism records the tokens and cost of every request. When the month's budget is spent, a rule can send everything to free local models.
Your subscription as a provider. A logged-in Claude Code or Codex CLI on the same machine can answer requests, billed to your plan.
A web dashboard. A playground, the model list, your rules, usage charts, gateway status and Projects, where conversations and an optional agent live.
Companions. A Linux desktop window to start the gateway and manage local models, an MCP server for Claude Code, and VS Code and browser extensions you can build from source.

How a request flows

text

your tool ──► DL Prism (127.0.0.1:8130) ──► Anthropic, OpenAI, DeepSeek, ...
 (base URL)        rules, failover,          or Ollama, vLLM, LM Studio
                   cost ledger

Every answer carries headers that say which provider and model answered and why: X-Prism-Provider, X-Prism-Model, X-Prism-Mode and X-Prism-Reason.

What stays on your machine

Prism listens only on 127.0.0.1 by default, so other computers cannot reach it.
API keys are read from environment variables and never appear in logs or API responses.
The usage record holds token counts and costs only, never your prompts.
Requests go only to the providers you configure. A setup with only local models works with no internet connection.

Who it is for

Developers who use several AI models and want one place to switch between them.
Anyone who wants to try a cheap or local model first and fall back to a bigger one.
People who want a hard monthly limit on AI spending.

Platforms

Part	Platforms
Gateway (`prism`)	Linux x64 and arm64, Windows x64 and arm64
Desktop companion	Linux x64 and arm64

The gateway is a single file with the dashboard built in. There is no macOS build.

DL Prism ​

What it does ​

How a request flows ​

What stays on your machine ​

Who it is for ​

Platforms ​