Memory search

C5 keeps an index of the notes, memories and documents your agents produce, so an agent can find what it — or another agent — wrote weeks ago without you pointing at the file.

Two kinds of search sit behind it:

• Keyword search finds documents containing the words you typed.
• Semantic search finds documents that mean the same thing, even when they use different words. Asking about "customer churn" can surface a note titled "why people cancel".

Keyword search works the moment C5 starts. Semantic search needs language models, and those are the one thing C5 does not ship inside the download.

The one-time model download #

The models come to about 2 GB. Bundling them would roughly quadruple the size of every C5 download to save a fetch that happens once, so C5 gets them the first time it starts instead.

You do not have to do anything. C5 downloads them in the background, keeps working while they arrive, and says so in its startup log:

[mcp] QMD: fetching embedding, reranking, generation model(s) in the background (~2 GB, one time)

When they land, semantic search turns on by itself. Until then C5 tells you what is still missing rather than reporting a fault:

[mcp] QMD query limited without download requirements

That line means C5 is running normally with keyword search, and the models have not arrived yet. Nothing is broken.

Where they go #

Everything lives inside your C5 directory rather than scattered through your system:

~/.growther/qmd/

On Windows that is %USERPROFILE%\.growther\qmd.

You will find a README.md in there listing every model, its size, and the exact path for your machine. It is written when C5 first starts, and C5 never overwrites your own notes in it.

If the machine has no internet #

On a restricted or air-gapped network, fetch the models somewhere that can reach the internet and copy them across:

1. Download all three from HuggingFace:

   • embedding — https://huggingface.co/ggml-org/embeddinggemma-300M-GGUF
   • reranking — https://huggingface.co/ggml-org/Qwen3-Reranker-0.6B-Q8_0-GGUF
   • generation — https://huggingface.co/tobil/qmd-query-expansion-1.7B-gguf

2. Rename each to the exact filename listed in ~/.growther/qmd/README.md and put them in the folder that file names.

   The name matters. C5 finds a model by its filename. The right file under the wrong name is ignored, and C5 tries to download it again.

3. Restart C5. There is nothing to configure.

You do not have to take all three. Each one adds something on its own — the embedding model alone enables semantic search.

Fetching them yourself #

To pull the models on your own schedule — before taking a laptop somewhere without signal, say — run:

growther qmd-run pull

To stop C5 fetching them automatically, start it with GROWTHER_QMD_NO_AUTO_DOWNLOAD=1.

Checking it #

growther qmd-run doctor

This reports which models C5 can see and whether the search index is healthy.

Windows on ARM #

Windows on ARM has no build of the component C5 uses for semantic search. On that platform keyword search, indexing and document access all work normally; semantic search does not. Every other platform has the full set.