Layer 5 · self-hosting reality check
What it actually takes to self-host Aider
The docs say Not stated by the vendor. Aider is a Python CLI and needs almost nothing when it calls a hosted API. Run the model locally and the model sets the floor: Qwen2.5-Coder, the family in aider's Ollama docs, is a 4.7 GB download at 7B and 20 GB at 32B (4-bit) in the Ollama library.. In practice you want With an API key: any laptop. Locally, to be usable: enough GPU or unified memory for a ~20 GB model plus its context (qwen2.5-coder:32b at 4-bit). In practice that means a 24 GB GPU or a 32 GB+ Apple Silicon Mac. The fp16 build named in aider's own config example is 66 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 Cursor alternatives, where Aider is one of the picks.
Wondering whether you need to at all? Is Cursor free? — what the free tier actually allows, and where the wall is.
| RAM — documented minimum | Not stated by the vendor. Aider is a Python CLI and needs almost nothing when it calls a hosted API. Run the model locally and the model sets the floor: Qwen2.5-Coder, the family in aider's Ollama docs, is a 4.7 GB download at 7B and 20 GB at 32B (4-bit) in the Ollama library. |
|---|---|
| RAM — what it really needs | With an API key: any laptop. Locally, to be usable: enough GPU or unified memory for a ~20 GB model plus its context (qwen2.5-coder:32b at 4-bit). In practice that means a 24 GB GPU or a 32 GB+ Apple Silicon Mac. The fp16 build named in aider's own config example is 66 GB. |
| CPU | Any modern CPU for aider itself. With a local model, the GPU sets the speed. |
| Disk | Under 1 GB for aider. Local models are 5–66 GB each, depending on size and quantisation. |
| Monthly cost | $0 — it runs on the machine you already own. The model is the bill: you pay per token for API use (aider's docs warn that adding extra files to the chat costs more tokens), or you need a GPU big enough for a local model. |
| Setup time | 10 minutes with an API key. An afternoon to get a local model behaving. |
| How you install it | `python -m pip install aider-install && aider-install` (or the curl / PowerShell one-liner), then set an API key. For local use, run Ollama and start `aider --model ollama_chat/<model>`. |
| Ongoing maintenance | Low: it is a CLI you upgrade now and then. Releases have slowed down. 0.86.2 (February 2026) is the newest on PyPI as of 30 Sep 2026, it needs Python 3.10–3.12, and the model list in the docs still names 2025-era models. |
| Where it stops scaling | One developer per install. There is nothing shared or centrally managed, so every developer on a team needs their own API key or local model, and costs grow with how many files each person adds to the chat. |
The thing that catches people out
Aider's own troubleshooting page says most local models are "just barely capable" of working with it, so edit errors are "probably unavoidable". Quantized models are worse, and those are the ones that fit a consumer GPU. When it goes wrong, the model writes code in the chat that aider can't apply to your files or commit. If you run local, use the biggest coding model your memory holds and try `--edit-format whole` or architect mode, as the docs suggest. Keep a strong API model set up for the edits that matter.
When not to self-host Aider
You want a free Cursor on a laptop without a big GPU: you'll either pay per token or fight a small model's edit errors. Skip it too if you want inline completions inside an IDE rather than a terminal chat that commits to Git. Cline or Continue fit that better.
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 5 self-hosting guides
- Self-hosting LibreOffice2 GB with a large spreadsheet open
- Self-hosting ONLYOFFICE6 GB for the Document Server with a handful of concurrent editors
- Self-hosting Collabora Online4 GB, and roughly 1 GB per 20 concurrent documents
- Self-hosting CryptPad2 GB for a small instance
- Self-hosting Mattermost4 GB for a team of 50 with PostgreSQL on the same box
- Self-hosting Rocket.Chat6 GB with MongoDB on the same machine
Common questions
- How much RAM does Aider actually need?
- With an API key: any laptop. Locally, to be usable: enough GPU or unified memory for a ~20 GB model plus its context (qwen2.5-coder:32b at 4-bit). In practice that means a 24 GB GPU or a 32 GB+ Apple Silicon Mac. The fp16 build named in aider's own config example is 66 GB. in practice. The documented minimum is Not stated by the vendor. Aider is a Python CLI and needs almost nothing when it calls a hosted API. Run the model locally and the model sets the floor: Qwen2.5-Coder, the family in aider's Ollama docs, is a 4.7 GB download at 7B and 20 GB at 32B (4-bit) in the Ollama library., which is the figure at which the process starts rather than the figure at which it works under real use. Any modern CPU for aider itself. With a local model, the GPU sets the speed. alongside it.
- What does self-hosting Aider cost per month?
- $0 — it runs on the machine you already own. The model is the bill: you pay per token for API use (aider's docs warn that adding extra files to the chat costs more tokens), or you need a GPU big enough for a local model. This is commodity VPS pricing and excludes your time, which is the larger cost for most people — budget for low: it is a CLI you upgrade now and then. Releases have slowed down. 0.86.2 (February 2026) is the newest on PyPI as of 30 Sep 2026, it needs Python 3.10–3.12, and the model list in the docs still names 2025-era models.
- How long does it take to set up Aider?
- 10 minutes with an API key. An afternoon to get a local model behaving., via `python -m pip install aider-install && aider-install` (or the curl / PowerShell one-liner), then set an API key. For local use, run Ollama and start `aider --model ollama_chat/<model>`..
- When should I NOT self-host Aider?
- You want a free Cursor on a laptop without a big GPU: you'll either pay per token or fight a small model's edit errors. Skip it too if you want inline completions inside an IDE rather than a terminal chat that commits to Git. Cline or Continue fit that better.
- What is the most common mistake when self-hosting Aider?
- Aider's own troubleshooting page says most local models are "just barely capable" of working with it, so edit errors are "probably unavoidable". Quantized models are worse, and those are the ones that fit a consumer GPU. When it goes wrong, the model writes code in the chat that aider can't apply to your files or commit. If you run local, use the biggest coding model your memory holds and try `--edit-format whole` or architect mode, as the docs suggest. Keep a strong API model set up for the edits that matter.