简体中文
EnglishCan I self-host Manifest, and what does that include?
Updated
Yes. Manifest is described on its site as an open source LLM gateway, and self-hosting is an explicit option: you can "run the whole LLM router on your infrastructure or your machine," and the page labels it "Free and open source." Self-hosting therefore covers the core routing layer — one endpoint to reach your models and providers, fallback and request repair, and request/token/cost visibility — rather than only a client SDK. The site presents self-hosting alongside a "Get Started for Free" path, so the practical choice is between signing up for the hosted product and running the router yourself; the page does not detail installation steps, supported deployment targets, or what (if anything) differs between the two paths.
What self-hosting includes
The page groups the gateway's capabilities under three functions, all of which belong to the router you would run:
- Route: a single endpoint to access all your models and providers, including OpenAI-compatible APIs, bring-your-own-key providers, custom providers, and local models.
- Fix: fallback to another provider when one is down, plus Autofix, which repairs broken requests before the error reaches your agent.
- Observe: tracking of every request, token, and dollar, with full request and response body logs and cost breakdowns by agent, key, and provider.
Because these are described as features of the gateway itself, self-hosting is the way to keep routing, logging, and cost tracking on infrastructure you control — for example, when prompts or responses should not leave your network, or when you want local models to stay local while only selected traffic goes to the cloud.
Self-hosting vs. the hosted signup
| Dimension | Self-host | Free signup (hosted) |
|---|---|---|
| Where the router runs | Your infrastructure or machine | Manifest's service |
| Cost model | Described as free and open source | Not specified on the page beyond a free start |
| Routing, fallback, Autofix | Part of the router | Part of the gateway |
| Logging and cost tracking | Part of the router | Part of the gateway |
| Setup effort | You operate it | No infrastructure work described |
The page does not state pricing for the hosted tier, whether login is required, or which features are gated. Treat "free and open source" as a statement about the self-hosted router, not as a claim that every hosted capability is free.
Where to start
- Decide on data boundaries. List which providers must be reached through your own network and which models can run locally (Ollama, LM Studio, llama.cpp, or any local server are named as options).
- Choose the path. If you need the router inside your infrastructure, take the self-host route; if you mainly want to try routing and cost tracking quickly, the free signup is the lower-effort entry.
- Connect credentials. Manifest accepts API keys from popular providers and OAuth tokens for subscription plans, so bring-your-own-key and subscription access both work through the same endpoint.
- Configure routes. Model and parameter settings are described as configurable per route without code changes, so you can set fallback order and per-route parameters visually.
- Verify. After routing traffic, confirm in the logs that requests, tokens, and costs are being recorded per agent, key, and provider — that is the check that observation is working as intended.
A common sticking point is assuming self-hosting means losing Autofix or fallback; the page presents both as gateway features, with Autofix available "through the LLM Gateway or standalone." Another is expecting the page to answer deployment questions — it doesn't cover container images, hardware requirements, or upgrade procedures, so plan to confirm those from the project's repository before committing.