cd /news/ai-products/manifest-is-taking-a-new-direction · home topics ai-products article
[ARTICLE · art-117770] src=manifest.build ↗ pub= topic=ai-products verified=true sentiment=· neutral

Manifest is taking a new direction

Manifest, the open-source AI gateway, is pivoting to become an autofix solution for failing API requests, not just LLM calls, after its Autofix feature reduced errors by 36% on live traffic for one client. The company will repair LLM, external API, internal API, and MCP requests in real time, applying patches from a knowledge base or proposing and testing new ones, with cloud and self-hosted options and a free tier up to a monthly volume.

read2 min views1 publishedSep 1, 2026
Manifest is taking a new direction
Image: Manifest (auto-discovered)

Hey everyone,

As some of you already saw, we have spent the last weeks focused on Autofix, the Manifest feature that repairs failing LLM requests while your app runs. The feedback convinced us to go further. Manifest is becoming an autofix solution, not only for LLM calls, but for APIs in general.

Why this change #

The interest we received on Autofix went beyond what we expected. And it points at a real problem: API errors cause downtime, and downtime costs companies money every year. Most tools attack this through codebase analysis, and API connections are their blind spot. A request that breaks because an API changed will pass every code review. We faced this ourselves: running our gateway showed us thousands of real production errors, and most of them turned out to be repairable. On one client platform, Autofix reduced errors by 36% on live traffic.

What Manifest will fix #

Manifest will repair calls that come back with an error:

  • LLM API requests
  • API requests to external services or internal APIs - MCP requests, the tool calls your AI agents send

For example, if your app calls an external service (Stripe, or any other) and the request is rejected because one field is no longer accepted, Manifest will read the error, remove the field and retry. The call will go through in less than a second, with no developer involved.

How it will work #

We will catch the failed request and analyze its error. If our knowledge base already holds the patch for that error, we will apply it. If it does not, we will propose a candidate patch, test it against a real call and create it once approved. Then we will send the repaired request again. Everything will happen live, so your agent stays up.

Autofix will run in the cloud. Companies that cannot let their data out will be able to run it self-hosted. Contact us if you have on-prem needs.
Contact us

It will be free up to a monthly volume of repaired requests.

What about the gateway? #

The gateway stays. It remains open source and maintained. And it will use the Autofix product itself.

Stay in the loop #

Leave your email and you will hear from us the day it ships.

If you want a live demo of Autofix on your internal APIs, request a demo and we will walk you through it.

── more in #ai-products 4 stories · sorted by recency
── more on @manifest 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
Live at https://your-agent.zahid.host
Get free account → Pricing
from €0/mo · no card required
LIVE [news/manifest-is-taking-a…] indexed:0 read:2min 2026-09-01 ·