cambriadaniele

Librechat setup

2026-04-12T00:00:00+00:00

Everything I need to not forget. Deployed to:

librechat.cambriadaniele.com</a>

If you're reading this, sorry, registration is deactivated.

Currently running: OpenAI, MiniMax, and Litellm endpoints on a $6/mo DigitalOcean droplet (1 vCPU, 1GB RAM).

On Digital Ocean, choose a droplet with Docker already installed. The $6/month plan (1 vCPU, 1GB RAM) works fine but requires memory management.

Follow the official guide</a>.

To start the app container, run:

docker compose up -d
</code></pre>
To stop the container, run:</p>
docker compose down
</code></pre>
Configuration</a></h2>
librechat.yaml</a></h3>
Mount your configuration file into the container via docker-compose.override.yml</code>:</p>
services:
  api:
    volumes:
      - type: bind
        source: ./librechat.yaml
        target: /app/librechat.yaml
</code></pre>
The configuration defines all AI providers and models available to users.</p>
.env File</a></h3>
Store all API keys in .env</code>. Required keys:</p>
OPENAI_API_KEY=sk-...
MINIMAX_API_KEY=sk-cp-...
LITELLM_API_KEY=...
</code></pre>
AI Providers</a></h2>
OpenAI (Direct)</a></h3>

Latest models: gpt-5.4, gpt-5.4-mini, gpt-5.4-nano, gpt-5.4-pro</li>
Realtime models: gpt-realtime-mini, gpt-realtime-1.5</li>
Vision: gpt-image-1-mini, gpt-image-1.5</li>
Reasoning: o3-deep-research, o4-mini-deep-research</li>
Legacy: gpt-4o, gpt-4o-mini</li>
</ul>
Configure in librechat.yaml</code>:</p>
endpoints:
  openAI:
    apiKey: "${OPENAI_API_KEY}"
    models:
      default:
        - "gpt-5.4-nano"
        - "gpt-5.4-mini"
        - "gpt-5.4"
        - "gpt-5.4-pro"
        # ... etc
</code></pre>
MiniMax</a></h3>
Chinese AI provider with strong code understanding and reasoning.</p>
custom:
  - name: "MiniMax"
    apiKey: "${MINIMAX_API_KEY}"
    baseURL: "https://api.minimax.io/v1"
    models:
      default:
        - "MiniMax-M2.7"
</code></pre>
Litellm</a></h3>
Proxy endpoint for multiple providers. Currently configured with:</p>

OpenAI models via Litellm</li>
Anthropic Claude</li>
Google Gemini</li>
Local models (Llama, Mistral, etc.)</li>
</ul>
- name: "Litellm"
  apiKey: "${LITELLM_API_KEY}"
  baseURL: "https://litellm.sph-prod.ethz.ch/v1"
  models:
    default:
      - "openai/gpt-5.4-mini"
      - "anthropic/claude-sonnet-4-5"
      - "gemini/gemini-1"
      # ... etc
</code></pre>
Memory Management</a></h2>
On 1GB droplet, set realistic memory limits in docker-compose.override.yml</code>:</p>
services:
  api:
    deploy:
      resources:
        limits:
          memory: 550M
        reservations:
          memory: 400M

  mongodb:
    deploy:
      resources:
        limits:
          memory: 250M
        reservations:
          memory: 200M

  meilisearch:
    deploy:
      resources:
        limits:
          memory: 100M

  rag_api:
    deploy:
      resources:
        limits:
          memory: 80M

  vectordb:
    deploy:
      resources:
        limits:
          memory: 70M
</code></pre>
Health Checks</a></h2>
Add health checks to auto-restart crashed containers:</p>
services:
  api:
    healthcheck:
      test:
        [
          "CMD-SHELL",
          "wget --quiet --tries=1 --spider http://127.0.0.1:3080/ || exit 1",
        ]
      interval: 30s
      timeout: 10s
      retries: 3
      start_period: 60s
</code></pre>
Manual Update Instructions (For Forked Repo with Upstream)</a></h2>
Fetch latest changes from the original repo.</p>
git fetch upstream
</code></pre>
Update the local main</code> branch</p>
git checkout main
git merge upstream/main
</code></pre>
Rebase your custom branch onto the updated main</code> branch to incorporate the latest changes.</p>
git checkout librechat-domain
git rebase main
</code></pre>


If conflicts occur (e.g., in nginx.conf</code>):</p>
# Resolve conflicts manually, then:
git add client/nginx.conf
git rebase --continue
</code></pre>
</li>
</ul>
Push updates to your forked repository. Use --force-with-lease</code> to avoid overwriting changes.</p>
git push origin main
git push --force-with-lease origin librechat-domain
</code></pre>

Updating Docker Containers</a></h2>
Stop containers (no data loss):</p>
docker compose down
</code></pre>
Pull latest images:</p>
docker compose pull
</code></pre>
Start LibreChat:</p>
docker compose up -d
</code></pre>
Verify the update:</p>
docker ps  # Check containers are running
curl -I http://localhost:80  # Or the domain
</code></pre>
If you get 502 errors after restart, NGINX may need to pick up new container IPs:</p>
docker restart LibreChat-NGINX
</code></pre>



Cheatsheet
2026-01-09T00:00:00+00:00
This cheatsheet is a work in progress.</p>
Download the PDF (Exam version)</a>
Download the PDF (study version)</a></p>


Cheatsheet
2026-01-09T00:00:00+00:00
This cheatsheet is a work in progress.</p>
Download the PDF</a></p>


Setup of TeXpresso for Neovim
2025-05-19T00:00:00+00:00
Everything I need to not forget.
This is my basic configuration for TeXpresso on Neovim, specifically Lazyvim.</p>
Video Example</a></h1>
From the TeXpresso repository</a>.</p>

  
  Your browser does not support the video tag.
</video>
Setup</a></h1>
Installation</a></h2>
Follow these instructions</a>.</p>
Configuration</a></h2>
In ~/.config/nvim/lua/plugins/texpresso.lua</code>:</p>
return {
  {
    "let-def/texpresso.vim",
    ft = "tex", -- load only for TeX files for efficiency
    config = function()
      -- Set the custom path to your texpresso binary
      require("texpresso").texpresso_path = "/home/yourname/dev/texpresso/texpresso"
    end,
  }
}
</code></pre>
In ~/.config/nvim/lua/config/keymamps.lua</code>:</p>
vim.keymap.set(
  "n",
  "<leader>öt", -- Or whatever you may want to use as a keymap
  ":TeXpresso %<CR>",
  vim.tbl_extend("force", opts, { desc = "Run TeXpresso on current file" })
)
</code></pre>
Controls</a></h2>
Keyboard controls:</a></h3>

←, →: change page</li>
↑, ↓: move within the page</li>
p (for "page"): switch between "fit-to-page" and "fit-to-width" zoom modes</li>
c ("crop"): crop borders</li>
q ("quit"): quit</li>
i ("invert"): dark mode</li>
I : toggle theming</li>
t ("top"): toggle stay-on-top (keeping TeXpresso above the editor window)</li>
b ("border"): toggle window borders</li>
F5: start fullscreen presentation (leave with ESC)</li>
</ul>
Mouse controls:</a></h3>

click: select text in window (TODO: move Emacs buffer with SyncTeX)</li>
control+click: pan page</li>
wheel: scroll page</li>
control+wheel: zoom</li>
</ul>


Setup of AI tools for Neovim
2025-05-14T00:00:00+00:00
Everything I need to not forget.</p>
CodeCompanion</a></h1>
Configuration</a></h2>
Standard Lazy.nvim configuration, with some extra options.</p>
{
  "olimorris/codecompanion.nvim",
  opts = {
    adapters = {
      deepseek = function()
        return require("codecompanion.adapters").extend("deepseek", {
          env = {
            api_key = "KEY", -- At the moment I left it hardcoded, but you can use vim.env.KEY or vim.fn.getenv("KEY") to get the key from the environment
          },
          schema = {
            model = {
              default = "deepseek-chat",
            },
          },
        })
      end,
    },
    display = {
      action_palette = {
        -- width = 95,
        -- height = 10,
        prompt = "Prompt ", -- Prompt used for interactive LLM calls
        provider = "default", -- Can be "default", "telescope", "mini_pick" or "snacks". If not specified, the plugin will autodetect installed providers.
        opts = {
          show_default_actions = true, -- Show the default actions in the action palette?
          show_default_prompt_library = true, -- Show the default prompt library in the action palette?
        },
      },
    },
    strategies = {
      chat = {
        -- adapter = "deepseek",
        -- adapter = "anthropic",
      },
      inline = {
        -- adapter = "copilot",
      },
      cmd = {
        -- adapter = "deepseek",
      },
    },
    extensions = {
      history = {
        enabled = true,
        opts = {
          -- Keymap to open history from chat buffer (default: gh)
          keymap = "gh",
          -- Automatically generate titles for new chats
          auto_generate_title = true,
          ---On exiting and entering neovim, loads the last chat on opening chat
          continue_last_chat = false,
          ---When chat is cleared with `gx` delete the chat from history
          delete_on_clearing_chat = false,
          -- Picker interface ("telescope" or "snacks" or "default")
          picker = "telescope",
          ---Enable detailed logging for history extension
          enable_logging = false,
          ---Directory path to save the chats
          dir_to_save = vim.fn.stdpath("data") .. "/codecompanion-history",
          -- Save all chats by default
          auto_save = true,
          -- Keymap to save the current chat manually
          save_chat_keymap = "sc",
        },
      },
      vectorcode = {
        opts = {
          add_tool = true,
        },
      },
    },
  },
  dependencies = {
    "nvim-lua/plenary.nvim",
    "nvim-treesitter/nvim-treesitter",
    "ravitemer/codecompanion-history.nvim",
  },
}
</code></pre>
Extensions</a></h2>
History</a></h3>
Just install the plugin:</p>
{
    "olimorris/codecompanion.nvim",
    dependencies = {
        --other plugins
        "ravitemer/codecompanion-history.nvim"
    }
}
</code></pre>
And add it to the extensions as shown above.
While in the CodeCompanion chat, use gh</code> to open the history, and sc</code> to save the current chat.</p>
VectorCode</a></h3>
Configuration is needed. First install pipx. Then install the package:</p>
pipx install vectorcode
</code></pre>
Then, cd</code> into your code and run vectorcode init</code> to initialize the vector database.
Finally, run:</p>
vectorcode vectorise src/**/*.py
</code></pre>
Also install the plugin:</p>
{
  "Davidyz/VectorCode",
  version = "*", -- optional, depending on whether you're on nightly or release
  build = "pipx upgrade vectorcode", -- optional but recommended. This keeps your CLI up-to-date.
  dependencies = { "nvim-lua/plenary.nvim" },
}
</code></pre>
To grant the LLM access to your indexed codebases, simply mention the @vectorcode</code> tool in the chat buffer. The LLM can then query any projects indexed by VectorCode (verifiable via vectorcode ls</code> in the terminal) to retrieve relevant context for your prompts.</p>


Digital Ocean Droplet setup
2025-04-05T00:00:00+00:00
Everything I need to not forget.</p>
Digital Ocean</a></h1>
Droplet Creation</a></h2>
Create a Droplet with Docker already installed</em>. Add an ssh key to the droplet, during creation already.</p>
To authenticate:
Go to GitHub → Settings → Developer Settings → Personal Access Tokens (Classic)
Create a token with repo permissions.
Configure git to use the token for authentication:</p>
git config --global credential.helper store
</code></pre>
And then clone a repository, and enter the username and PAT.</p>
Neovim, Lazyvim</a></h2>
curl -LO https://github.com/neovim/neovim/releases/latest/download/nvim-linux-x86_64.appimage
chmod u+x nvim-linux-x86_64.appimage
./nvim-linux-x86_64.appimage

mkdir -p /opt/nvim
mv nvim-linux-x86_64.appimage /opt/nvim/nvim
</code></pre>
Then:</p>
nano ~/.bashrc
# Add this:
export PATH="$PATH:/opt/nvim"
# Save with CtrlO and close with CtrlX
source ~/.bashrc
</code></pre>
Then:</p>
mv ~/.config/nvim{,.bak}
git clone https://github.com/LazyVim/starter ~/.config/nvim
rm -rf ~/.config/nvim/.git
nvim
</code></pre>
Github Actions</a></h1>
Secrets</a></h2>
Go to Settings -> Secrets and variables -> Actions -> New repository secret</code>, and add the following secrets:</p>

DIGITAL_OCEAN_DROPLET_IP</li>
DIGITALOCEAN_SSH_PRIVATE_KEY</li>
DIGITALOCEAN_SSH_USER (root for personal use, otherwise create a new user)</li>
SSH_PASSWORD (If the local ssh key is password protected, otherwise leave it empty)</li>
</ul>
Action</a></h2>
Whenever something gets pushed to main, it gets pulled to opt/thesis in the droplet.
I used this action for some containers I was running for data-scraping for my thesis, hence the directory name.
It then builds and gets the compose containers up, with environment variable PRODUCTION_ENV = True.</p>
name: Deploy to DigitalOcean

on:
  push:
    branches: [main]

jobs:
  deploy:
    runs-on: ubuntu-latest

    steps:
      - name: Checkout code
        uses: actions/checkout@v4

      - name: Deploy via SSH
        uses: appleboy/ssh-action@v1.0.0
        with:
          host: ${{ secrets.DIGITALOCEAN_DROPLET_IP }}
          username: ${{ secrets.DIGITALOCEAN_SSH_USER }}
          key: ${{ secrets.DIGITALOCEAN_SSH_PRIVATE_KEY }}
          passphrase: ${{ secrets.SSH_PASSWORD }}
          script: |
            cd /opt/thesis
            git pull origin main
            PRODUCTION_ENV=True docker compose up -d --build
</code></pre>


Summary
2024-10-19T00:00:00+00:00
This summary is a work in progress. As the course has already finished, I don't know if I will ever complete it.</p>
Download the PDF</a></p>


Website setup
2024-10-18T00:00:00+00:00
Everything I need to not forget.</p>
Zola</a></h1>
Themes</a></h2>
After choosing a theme:</p>
zola init mysite
git init  
git submodule add https://github.com/username/theme_name.git themes/theme_name
git submodule update --init --recursive   #very important
</mark>git submodule update --remote --merge     #very important
</mark></code></pre>
Then create/import some content, copy the templates and eventually the rest.
Check if it runs with</p>
zola serve 
</code></pre>
If using a custom domain, add a CNAME</code> file to static/</code>, with the domain name written inside.</p>
Github</a></h1>
Repository</a></h2>
To serve a site at username.github.io</code>, name the repository <username>.github.io</code>,
(otherwise GitHub will append the repository name to the URL, e.g.: username.github.io/repositoryname</code>)</p>
In the repository setting, activate Actions -> General -> Workflow permissions -> Read and write permissions</code>,
otherwise the Github Action trying to build the website will fail.</p>
Once the repository has been created, stage and commit the local changes, add the remote origin and push the changes.</p>
Github Actions</a></h2>
Go to Actions -> new workflow -> setup a workflow yourself</code>, and create a .yml</code> file:</p>
on: push
name: Build and deploy GH Pages
jobs:
  build:
    runs-on: ubuntu-latest
    if: github.ref == 'refs/heads/main'
    steps:
      - name: checkout
        uses: actions/checkout@v4
      - name: build_and_deploy
        uses: shalzz/zola-deploy-action@v0.19.2
        env:
          # Target branch
          PAGES_BRANCH: gh-pages
          # Provide personal access token
          TOKEN: ${{ secrets.GITHUB_TOKEN }}
</code></pre>
Commit and push the changes, wait for the job to terminate, and as soon as the gh-pages branch has been created,
go to Settings -> Pages -> Deploy from branch -> gh-pages</code>, and also check if the custom domain has been correctly
setup.</p>


Problem Set 1
2022-01-01T00:00:00+00:00
Documentation</a></h3>
The documentation for simple_turing_machine can be found here</a>.</p>

cambriadaniele

Librechat setup

AI Providers</a></h2> OpenAI (Direct)</a></h3> Latest models: gpt-5.4, gpt-5.4-mini, gpt-5.4-nano, gpt-5.4-pro</li> Realtime models: gpt-realtime-mini, gpt-realtime-1.5</li> Vision: gpt-image-1-mini, gpt-image-1.5</li> Reasoning: o3-deep-research, o4-mini-deep-research</li>

OpenAI (Direct)</a></h3> Latest models: gpt-5.4, gpt-5.4-mini, gpt-5.4-nano, gpt-5.4-pro</li> Realtime models: gpt-realtime-mini, gpt-realtime-1.5</li> Vision: gpt-image-1-mini, gpt-image-1.5</li> Reasoning: o3-deep-research, o4-mini-deep-research</li>

Litellm</a></h3> Proxy endpoint for multiple providers. Currently configured with:</p> OpenAI models via Litellm</li> Anthropic Claude</li> Google Gemini</li>

Cheatsheet

Cheatsheet

Setup of TeXpresso for Neovim

Setup of AI tools for Neovim

Digital Ocean Droplet setup

Summary

Website setup

Problem Set 1