How I Fit Run Aider or Goose for VPS

Why person dey run coding agent for VPS

Coding agent na tool wey dey read your codebase, write changes, and run commands to test dem, all from terminal. To run one for VPS get three advantages pass laptop: e fit continue work after you disconnect, e dey close to your code and build tools, and e fit talk to model wey you host by yourself. Start am inside tmux session, then you fit close your laptop and continue the work later, just like running Claude Code for VPS with tmux.

Two terminal-based agents dey worth knowing. Aider na terminal-native, git-first option: e dey auto-commit every change with sensible message, so your history remain clean and every edit fit reverse. E dey among the terminal coding agents wey don dey around for longest time, though e no dey release new versions as fast again. Goose, from Block, get wider scope: na Apache-2.0 agent wey get plenty extensions and support many providers, including local Ollama models. Linux Foundation's Agentic AI Foundation (AAIF) dey develop am now. Aider fit disciplined, git-centric workflow well; Goose fit general assistant wey get plenty plugins.

Wetin you need

You need VPS wey get git, tmux, and recent Python for Aider, or Goose installer for Goose. You also need model: either hosted API key from provider like Anthropic or OpenAI, or local model wey Ollama dey serve for the same VPS. Self-hosted option keep your code for your own server and e no cost anything per token, but you need enough memory for model wey fit work well.

Install and run Aider

Install Aider for an isolated environment with pipx. apt go provide pipx because Ubuntu 24.04 dey block plain pip install outside virtual environment (PEP 668). Then run Aider inside your project, for tmux session so e fit survive disconnection:

sudo apt install pipx
pipx ensurepath
pipx install aider-chat
tmux new -s aider
cd ~/my-project
aider

One version note: this installation dey work as e show for Ubuntu 24.04. For Ubuntu 26.04, e dey fail currently because Aider's pinned dependencies include older numpy wey no fit build against Ubuntu 26.04 Python 3.14. To fix am, give Aider im own Python 3.12:

pipx install --python 3.12 --fetch-missing-python aider-chat

pipx dey download standalone interpreter wey Aider alone go use, and installation go complete. This na symptom of Aider's slower release pace. If this one dey worry you, Goose na the option wey dem dey maintain more actively.

After e start, detach from the session with Ctrl-b then d. The agent go continue to work even when you close your laptop. Reattach later with tmux attach -t aider to read wetin e change. You fit keep another tmux window open for another agent. If both of dem na Claude Code, the two sessions fit pass work between themselves instead of you relaying am by hand. Aider dey read your repository, propose edits, apply dem, and commit each one. E fit also bring extra files for context when one change affect several files. Because every change na commit, to undo the agent na simple git revert. This na the safety net wey make Aider comfortable to use. Goose dey install with the documented one-line script and e dey run similarly from terminal, driving tasks through im extensions:

curl -fsSL https://github.com/aaif-goose/goose/releases/download/stable/download_cli.sh | bash

Model wey provider host or model wey you host by yourself

Hosted model dey give you the best quality and e no need local hardware, but you dey pay money for each token and your code dey go the provider. Cost for each token fit add up pass wetin most people expect, because the agent dey send the whole conversation again for every turn. The Claude Code explanation of wetin long agent session really dey spend for tokens apply here too, because Aider and Goose dey charge the same way. Self-hosted model through Ollama keep everything for your server and e free to run after you don pay for the hardware, but coding model wey good enough to help needs serious memory. Na this trade-off you need weigh: quality and convenience against privacy and cost.

If you choose self-hosted, the practical question na which model fit your server. Monitor the context window too, because Ollama dey default to small one and silently truncate anything wey long pass am. For coding agent, this mean files wey e just read fit drop out of the prompt: increase num_ctx go fix am, and the bigger KV cache wey e need na part of the memory wey you need budget. Size am before you download eight-gigabyte model go machine wey no fit hold am:

Safety: how e dey edit files and run commands

Coding agent no be passive. E dey rewrite files and fit run build and test commands, so give am the same care wey you go give any process wey fit change your system. Three habits fit cover most of the risk. Work inside git repository, so every change dey tracked and reversible, and Aider dey do this automatically for you. Run the agent as your ordinary unprivileged user, never as root, following users with least privilege, so bad command no fit touch the whole system. Also harden the box itself, because coding VPS still be public server: key-only SSH, default-deny firewall, and the rest. The OpenClaw hardening guide get the same approach for more autonomous agent, and the principles still apply.

To connect coding agent to your own workflow, how to build an AI agent with Claude show how one model dey drive your tools; how to run Google's Gemini CLI on a VPS na another terminal option, while self-hosting OpenHands na the heavier autonomous route. The open project wey get the most stars for this area now get guide of its own: how to run OpenCode on a VPS set am up with the same tmux and unprivileged-user pattern.

FAQ

I fit run coding agent with local model instead of paid API?

Yes. Both Aider and Goose dey work with local models wey Ollama serve, so you fit run coding agent fully self-hosted without per-token cost and without any code leaving your server. The catch na memory: model wey strong enough to write useful code need plenty RAM or VRAM, so size the machine based on the model before you commit.

Aider or Goose, which one I suppose use?

Choose Aider if your workflow dey terminal- and git-centric and you want the most mature option with the lowest overhead; e dey auto-commit every change, so your history remain reversible. Choose Goose if you want broader assistant with large extension ecosystem and support for many providers. Both dey run from terminal and both dey work with Ollama, so either one fit VPS.

How much memory I need for self-hosted coding model?

E depend on the model size and how much dem quantize am. Small quantized model fit run with few gigabytes, while stronger model need much more, and long context windows add to the memory use. Use the sizing tool above to estimate the memory wey particular model and context length need before you download am.

E safe to allow AI agent edit my code and run commands?

You fit manage am with correct habits. Keep your work inside git repository so every edit become reversible commit, run the agent as unprivileged user instead of root, and harden the VPS as you go do for any public server. Review the changes wey e commits instead of trusting dem blindly, especially any command wey e want run against your system.