aieph

MCP serverAI & models

Shared knowledge cache for AI coding agents — reuse an answer once it exists.

Available today. Use it from your connected AI after setup.

Connect ahel once, and every AI you use reads what you have installed.

Then ask your AI: use the aieph search tool from aieph

From the project's README

As published by tryaieph/aieph in README.md.

your fix, everyone's gain

aieph is a shared answer cache for AI coding assistants.

When an AI coding assistant is about to search the web, a small hook can quietly ask: "Does someone already have the answer?"

If the answer is already in the shared cache, aieph can return it right away — avoiding another trip to the web. If there is no answer, aieph simply gets out of the way.

MISS, timeout, or error? Nothing breaks. The original request continues as usual. That is the whole idea.

Your fix can become everyone's gain.

Why a shared cache?

AI coding assistants often look up the same kinds of things:

  • a familiar error message
  • a library's behavior
  • a configuration detail
  • a solution someone has already found

The first person does the searching. The next person doesn't necessarily need to. aieph gives those answers a place to be shared, and as more people use it, the cache can gradually become more useful.

It's still early, so a cache hit is not guaranteed. Sometimes aieph simply won't have the answer yet — and that's perfectly fine.

How it feels

AI assistant
     │
     │  "I'm about to search..."
     ▼
   aieph
     │
     ├── HIT  → answer is already here
     │
     └── MISS / timeout / error
              │
              ▼
          original search

aieph is designed to stay out of the way. There is no need to change how you work when the cache doesn't have something useful.

Getting started with Claude Code

Claude Code can use aieph with a single PreToolUse hook for WebSearch and WebFetch. A small setup looks like this:

git clone https://github.com/tryaieph/aieph.git

Then, in ~/.claude/settings.json, the relevant hook can look like:

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "WebSearch|WebFetch",
        "hooks": [
          {
            "type": "command",
            "command": "node /path/to/aieph/hook/aieph-cache.mjs"
          }
        ]
      }
    ]
  }
}

That's it.

Requirements: Node.js 18+. No account or API key is required.

Optional environment variables:

AIEPH_API_BASE
LOOKUP_TIMEOUT_MS

Cursor support is being prepared.

Removing aieph

If aieph is no longer useful for you, removing the PreToolUse block is enough to return to the original setup.

What gets sent?

aieph is intentionally simple and transparent. Only the search query text is sent for matching. Nothing else is sent. There is no account to create and no personal profile to maintain. When there is no matching answer, the request passes through normally.

Local shared memory

aieph also includes a small local-only memory CLI in cli/. It is separate from the shared answer cache. The idea is simple: write a fact once, and let connected assistants remember it later. This memory stays completely local.

Early days, honest expectations

aieph is still growing. The shared cache is young, so there will be plenty of moments when it doesn't have an answer yet. That's intentional: aieph never needs to be perfect to be useful. A miss should feel exactly like there was no cache at all.

Over time, every useful answer someone discovers has the potential to help the next person.

your fix, everyone's gain.

License

MIT

Tools it offers (1)

What this server listed when ahel dialed its public endpoint in Sep 2026, with no key and no account of yours. The names are the server’s own.

  • aieph_search

Signals

GitHub stars
1
Last commit
Sep 2026
Advanced
Delivery
aieph MCP server → your ahel gateway (mcp.ahel.ai) → every connected AI client.
Catalog kind
mcp-server
Gateway key
dev-aieph-aieph
Source
github.com/tryaieph/aieph
Hosted endpoint
https://aieph.dev/mcp