Is there a free plan on Manifest Cloud?

No. Manifest Cloud is $19/month with no request quota. The self-hosted gateway is free, unlimited and gets every feature Cloud gets.

Can we run Manifest on our own servers?

Yes. It is MIT-licensed and ships as a Docker image, so it runs on your own machines with no request limit and no data leaving your network. Enterprise adds on-prem/VPC deployment with SSO, audit logs and a support contract.

Which errors does Autofix actually repair?

Only request-side 4xx errors that a rewritten body can fix: 400 (bad parameter), 404 (moved model name) and 422 (rejected schema). It deliberately skips 401, 403, 429 and every 5xx — a bad key, rate limit or outage goes straight to fallback.

What data does Autofix send to the healing service?

The failed request body including message content, plus the provider's error response after secret scrubbing. API keys and OAuth tokens travel in headers and are never included. Nothing is sent for successful requests, and nothing at all while the toggle is off.

What happens if the healing service is down or slow?

Failures degrade quietly: your agent gets the original provider error, never a gateway 500. After three consecutive transport failures the gateway stops calling the healer for 30 seconds, and the boot health check is fire-and-forget so startup never blocks.

Can Autofix fix streaming requests?

Only if they fail before the first byte reaches your client. Once bytes are on the wire there is nothing to rewrite, so streaming failures mid-response fall through to your fallback chain instead.

Directory/Automation/AI Infrastructure/Manifest

Manifest

Manifest is an MIT-licensed LLM gateway that gives you one OpenAI-compatible endpoint across every provider, including local models via Ollama, LM Studio and llama.cpp. Bring your own keys or OAuth subscription tokens, set routing rules and model fallbacks visually, and keep full control over what goes to the cloud versus what stays on your machine. Its headline feature is Autofix: roughly 5% of LLM API calls fail in production, and most of those are request-side 4xx errors that a fallback model would reject too. Manifest intercepts those failures, asks a healing service for a corrected request body, and retries exactly once — before fallback kicks in. It deliberately skips 401, 403, 429 and all 5xx errors, so bad keys and outages still route to your fallback chain. Built for AI engineers, platform teams and founders shipping agents, it adds full request/response body logs, per-agent and per-provider cost tracking, budget alerts and error catalogs. Self-host the Docker image for free

Automation

Visit website

Added on

Nov 4, 2026

OpenAI compatibleAutofixModel fallbacksBring your own keyLocal modelsFull body logsCost trackingSelf-hosted DockerOAuth subscriptions

Product information

Everything worth knowing about Manifest

The complete picture — from everyday features to the technical detail builders and IT folks dig for.

Core features

What it actually does

Autofix

Broken LLM requests get repaired on the fly before your agent ever sees the error. The gateway sends the failed request plus the provider's error to a healing service, gets a corrected body back, and replays it once.

Model fallbacks

Automatically retry on another provider when one goes down or returns an error the request can't survive.

Bring your own key

Use your own API keys for every provider plus OAuth tokens for subscription plans (Claude, ChatGPT, Gemini). Manifest never marks up your token spend — you pay providers directly.

Local models

Route to Ollama, LM Studio, llama.cpp or any local server, and decide per route what goes to the cloud and what stays on your machine.

Cost & usage observability

See every dollar spent broken down by agent, API key and provider, with success rate, recovered requests and recovered-by-Autofix cards over any time range.

Full body request logs

Inspect the complete request and response body for every provider attempt — the failed original and the patched retry are logged separately so you can diff them side by side.

Technical capabilities

Scoped self-healing pipeline

Autofix only touches request-side 4xx errors (400, 404, 422) and deliberately skips 401, 403, 429 and all 5xx. Exactly one retry, 10s default healing timeout, three consecutive transport failures trigger a 30s circuit breaker, and credentials in headers are never sent to the healing service.

Must watch videos

Introducing World's 1st AI Commerce Agents Marketplace | Bik.ai | Manifest AI

1:39

Tutorial1:39

Introducing World's 1st AI Commerce Agents Marketplace | Bik.ai | Manifest AI

Introducing the World’s First AI Commerce Agents Marketplace No need to onboard 20 tools or hire 5 new people to scale your store. No tools. No humans. Just spin off agents — from 100+ ready-to-use or build your own inside the Agent Studio Forget paying for separate tools for 👇 🧩 Size Guides 🔍 Product Comparisons 📦 Order Tracking 💬 Support & Retention Just decide what you want to improve — ✨ Conversions 🔁 Repeat Purchases 💬 Support Efficiency Then simply: 1️⃣ Give a Goal 2️⃣ Add Instructions 3️⃣ Define Success Criteria And voilà — your agent is live ⚡️ Train once → deploy everywhere (WhatsApp, Instagram, Email, Messenger, Website). Your new teammates never sleep. Want to get started? ✅ No contracts ✅ No onboarding cost ✅ Begin with 1 agent — scale as you grow ✅ Setup: 100% white-glove service — my team trains your AI and takes it live See in action: https://getmanifest.ai/ai-commerce-agents #aiagents #ecommerceagents #ecommerceai #aiforecommerce #aiagent #ecommerceagent #aishoppingassistant #ordertracking #returnagent #shopifyagents

SNAP - Manifest (Ai Arabesk Cover) FULL ŞARKI

2:48

Tutorial2:48

SNAP - Manifest (Ai Arabesk Cover) FULL ŞARKI

Manibesk - SNAP (AI Arabesk Cover) Şarkı Sahibi: Manifest Söz & Müzik: Ateş Atilla Düzenleme: Suno AI Konuk Oyuncu: Ahmet Emre Dursun Videolarımı kaçırmamak için bana Youtubue'da abone olabilirsin. Ayrıca, video düzenleme, içerik üretimi ve yaratıcı süreçlerle ilgili daha fazla ipucu ve fikir için Instagram’da da beni takip edebilirsiniz. www.instagram.com/evliyacelebi Kullandığım Tüm Ürünler: (affiliate) https://app.hb.biz/l6CB3pcEC3os Masamdakiler: (affiliate) https://app.hb.biz/8OObu2ZUYBrR Işık Önerilerim: (affiliate) https://app.hb.biz/KwEAT6tAukWk

minefest - TRALALA (manifest) (AI Cover)

3:42

Tutorial3:42

minefest - TRALALA (manifest) (AI Cover)

Who uses it

Real use cases, no hype

Zero-Downtime Agents

Patch malformed requests on the fly so broken bodies never surface to your users.

Provider Failover

Fall back across providers automatically when one goes down or rate-limits you.

Cost Observability

Track every token and dollar by agent, key and provider in one dashboard.

BYOK & Subscriptions

Route API keys and OAuth subscription plans through a single endpoint.

Hybrid Local/Cloud Routing

Send sensitive traffic to Ollama or LM Studio and the rest to the cloud.

Platform Reliability

Cut avoidable client errors that churn users and burn revenue.

Private Self-Hosting

Run the whole gateway inside your VPC with no data leaving the network.

What's great

Autofix repairs broken 4xx requests before your agent sees them
MIT-licensed with free unlimited self-hosting via Docker
Zero token markup — bring your own keys and pay providers direct
One OpenAI-compatible endpoint for cloud, subscription and local models
Clean dashboard for cost, latency and failure tracking

Technical strengths

Autofix sits in front of the fallback chain, not instead of it
Circuit breaker skips the healer for 30s after 3 transport failures
Full request/response bodies logged per provider attempt

Where it falls short

Cloud has no free tier — $19/month minimum
Autofix ships failing request bodies to a hosted service
No repair once streaming bytes hit the wire
Cloud is single-user; seats and SSO lock behind Enterprise

Technical limitations

Autofix ignores 401/403/429 and all 5xx — no rewrite can save those
Repairable statuses limited to 400/404/422 by default
Self-hosters own upgrades, DB and uptime with no vendor SLA

For technical folks

The deep-dive specs

Deployment

Docker image; MIT-licensed, self-hostable on your own infrastructure

API compatibility

OpenAI compatible single endpoint

Autofix eligible statuses

400, 404, 422 (401/403/429 and all 5xx skipped)

Autofix timeout

10000 ms per healing call

Local model servers

Ollama, LM Studio, llama.cpp

Auth methods

Provider API keys and OAuth tokens for subscription plans

FAQ

Includes technical Q&A

No. You bring your own provider keys and pay providers directly. Manifest does not mark up tokens on any plan.

Traffic Insights

Monthly Visits

—

November 2026

Growth Rate

vs last month

Dominance

in Automation

Top Country

—

Top Countries Breakdown

—

Growth Trend

Stable

Traffic has remained stable.

Honest pricing

No sneaky tiers, no “contact sales”

Here's who each plan is actually for — and where the hidden charges might hit.

Self-hosted

Free

Self-hosters

Unlimited routed requests; All providers
including local models; Subscription providers (Claude
ChatGPT
Gemini...); Autofix
routing rules and fallbacks; Budget alerts and notifications; Your data stays on your machine; Community support via Discord

Manifest Cloud

Most picked

$19/mo

Teams

Hosted at app.manifest.build
ready in a minute; Updates land the same day we ship them; Servers
database and uptime handled by us; Unlimited routed requests; 365 days of dashboard history; Support from the team (platform issues
billing)

View Full Plans & Pricing on Manifest

Official pricing, checkout, and plan upgrades are handled directly on manifest.build.

Reviews