Layer 4 · self-hosting reality check
What it actually takes to self-host Aim
The docs say 512 MB. In practice you want 2 GB. Here is the honest version — real requirements, real monthly cost, what you will be maintaining, and the one thing that catches people out.
Usually reached from Weights & Biases alternatives, where Aim is one of the picks.
| RAM — documented minimum | 512 MB |
|---|---|
| RAM — what it really needs | 2 GB |
| CPU | 2 vCPU |
| Disk | Run metadata and metrics — modest unless you log images |
| Monthly cost | $12/mo, against per-seat experiment-tracking pricing |
| Setup time | 30 minutes |
| How you install it | pip install aim, then `aim up` — a single service, unlike MLflow's server-plus-database |
| Ongoing maintenance | Low. |
| Where it stops scaling | Genuinely fast at thousands of runs, which is exactly where MLflow's UI starts to struggle. |
The thing that catches people out
The repository is a local directory by default, which means the tracking server and your training jobs must share a filesystem — fine on one machine, broken the moment training moves to a cluster. Plan for remote tracking before you distribute training, not after your runs stop appearing.
When not to self-host Aim
You need a model registry or deployment tooling. Aim is deliberately tracking only; MLflow covers more ground.
Every guide here carries this section. A site that only ever tells you to self-host is selling something — the useful answer is sometimes no.
Other Layer 4 self-hosting guides
- Self-hosting Ollama8 GB VRAM for a 7B model at usable speed; 24 GB for 30B-class
- Self-hosting vLLM24 GB VRAM minimum for useful production serving
- Self-hosting QdrantVectors × dimensions × 4 bytes, in RAM, plus overhead — 1M × 768d is roughly 3 GB
- Self-hosting pgvector8 GB — the HNSW index wants to be resident
- Self-hosting LlamaIndex4 GB for the app; your vector store is the real cost
- Self-hosting faster-whisper5 GB VRAM for large-v3 in float16; 2 GB with int8
Common questions
- How much RAM does Aim actually need?
- 2 GB in practice. The documented minimum is 512 MB, which is the figure at which the process starts rather than the figure at which it works under real use. 2 vCPU alongside it.
- What does self-hosting Aim cost per month?
- $12/mo, against per-seat experiment-tracking pricing This is commodity VPS pricing and excludes your time, which is the larger cost for most people — budget for low.
- How long does it take to set up Aim?
- 30 minutes, via pip install aim, then `aim up` — a single service, unlike MLflow's server-plus-database.
- When should I NOT self-host Aim?
- You need a model registry or deployment tooling. Aim is deliberately tracking only; MLflow covers more ground.
- What is the most common mistake when self-hosting Aim?
- The repository is a local directory by default, which means the tracking server and your training jobs must share a filesystem — fine on one machine, broken the moment training moves to a cluster. Plan for remote tracking before you distribute training, not after your runs stop appearing.