[{"data":1,"prerenderedAt":161},["ShallowReactive",2],{"seo-verification":3,"blog-github-copilot-ai-credits-take-back-control-with-a-vps-en":6},{"google":4,"bing":5},"EycwPY2XMyTkVzas3n1ygeNJFGAH513qrMjfDljzsMQ","",{"id":7,"slug":8,"slugs":9,"title":12,"excerpt":13,"readTime":14,"views":15,"isPinned":16,"publishedAt":17,"category":18,"categories":24,"featuredImage":26,"bgImage":27,"posterImage":28,"relatedSolution":29,"intro":32,"sections":33,"ctaTitle":106,"ctaBody":107,"ctaButton":108,"ctaUrl":109,"relatedPosts":110},271,"github-copilot-ai-credits-take-back-control-with-a-vps",{"fr":10,"en":8,"ar":11},"github-copilot-ai-credits-alternative-vps","github-copilot-ai-credits-استعد-السيطرة-مع-خادم-vps","GitHub Copilot AI Credits: take back control with a VPS","Since June 2026, Copilot bills in AI Credits and predictability is gone. Host Ollama or Tabby on a VPS for a fixed-cost coding assistant with no prompt leakage.",8,0,false,"2026-08-16T00:00:00+00:00",{"id":19,"name":20,"slug":21,"color":22,"icon":23},1,"Artificial Intelligence","intelligence-artificielle","bg-purple-500\u002F10 text-purple-400","ia",[25],{"id":19,"name":20,"slug":21,"color":22,"icon":23},null,"\u002Fblog\u002Fcovers\u002Fbg.svg","\u002Fblog\u002Fcovers\u002Fgithub-copilot-ai-credits-alternative-vps-poster.svg",{"categorySlug":30,"appSlug":31},"artificial-intelligence","ollama","Since June 1, 2026, GitHub Copilot bills in AI Credits — units tied to token consumption. On the first day, Pro+ developers consumed 8 % of their 7 000-credit monthly quota in two hours during ordinary agentic sessions. Predictability vanished at exactly the moment usage was climbing. A direct alternative exists: host your own inference engine on a VPS, connect VS Code to it via Continue.dev, and get back to a fixed monthly cost without a single prompt leaving your server.",[34,38,48,51,54,76,79,94,97,100,103],{"type":35,"title":36,"body":37},"h2","Why the shift to AI Credits changes the equation","Until May 31, 2026, Copilot Pro+ charged a flat rate per month regardless of request volume. Since June 1, every Copilot plan includes a monthly AI Credits budget converted to tokens according to the model used. The result measured by early users: a two-hour agentic session consumed 8 % of the monthly Pro+ quota, with the 7 000-credit allotment estimating to run out in under two days at that pace. The problem is not the price per request — it is the absence of a floor. An agentic workflow that opens several files, calls tools and iterates consumes a multiple of what a simple completion uses. The predictability that was the promise of a flat fee no longer exists.",{"type":39,"title":40,"items":41},"ul","Six reasons to switch to a self-hosted coding assistant",[42,43,44,45,46,47],"**Fixed monthly cost**: a VPS running Ollama costs the same amount every month, regardless of agentic load.","**Prompt confidentiality**: no line of code, no project context leaves your infrastructure.","**Free model choice**: Qwen2.5-Coder, DeepSeek Coder, Llama 3.2 or any GGUF-format model available through Ollama.","**No per-session quota**: no degradation or blocking at the end of the month.","**Portability**: the same Ollama server feeds VS Code, JetBrains, Neovim or a CI pipeline.","**Independence from pricing decisions**: GitHub price or policy changes no longer affect your coding assistant.",{"type":35,"title":49,"body":50},"Concrete requirements","To run a 7B model quantised at Q4_K_M — the recommended format for a good quality\u002Fperformance balance — budget **at least 4 GB of RAM dedicated to the Ollama process**. With 8 GB, the model stays in memory between requests and avoids reloading. For 14B models, target 16 GB.\n\nOn the software side: Docker and Docker Compose installed on the VPS, a subdomain (e.g. `ollama.your-domain.com`) pointing to the VPS IP, and port 443 open. No GPU required: a 7B model in Q4_K_M runs on CPU at 5 to 15 tokens per second depending on the number of vCPUs, enough for function completions and interactive chat.",{"type":35,"title":52,"body":53},"Option 1 — Continue.dev + Ollama: setup in five minutes","Continue.dev installs like any VS Code extension. It reads a `config.json` file to know which model to connect to. That model can be local or remote — the remote case is what matters here: your VPS hosts the inference engine, your workstation hosts the IDE. The full chain: VS Code extension → `~\u002F.continue\u002Fconfig.json` → HTTPS URL of the reverse proxy → port `11434` of Ollama on the VPS.",{"type":55,"title":56,"steps":57},"steps","Deploy Continue.dev + Ollama on a VPS",[58,61,64,67,70,73],{"title":59,"body":60},"Provision the VPS and install Docker","Connect via SSH, update the system, then install Docker with the official installer:\n```bash\ncurl -fsSL https:\u002F\u002Fget.docker.com | sh\n```\nVerify with `docker compose version`. Create a dedicated folder: `mkdir -p \u002Fopt\u002Follama`.",{"title":62,"body":63},"Start Ollama and download a model","Launch the official container exposing the port only on localhost:\n```bash\ndocker run -d --name ollama -p 127.0.0.1:11434:11434 -v ollama:\u002Froot\u002F.ollama ollama\u002Follama\n```\nDownload a code-focused model: `docker exec ollama ollama pull qwen2.5-coder:7b-instruct-q4_K_M`. Download takes a few minutes depending on connection speed.",{"title":65,"body":66},"Configure the reverse proxy with TLS","Never expose port 11434 directly on the Internet — Ollama has no native authentication. With Caddy, one directive in the Caddyfile terminates TLS and relays to `localhost:11434`. Add basic authentication or a shared secret header to restrict access. Close the direct port: `ufw deny 11434`.",{"title":68,"body":69},"Install Continue.dev in VS Code","Open the Extensions panel (`Ctrl+Shift+X`), search for **Continue**, click Install. The extension adds an icon in the sidebar. Open `~\u002F.continue\u002Fconfig.json` (macOS\u002FLinux) via the gear icon in the Continue panel.",{"title":71,"body":72},"Edit config.json to point to the VPS","In the `models` array, add an ollama entry with the `apiBase` field pointing to the HTTPS URL of your reverse proxy. Fill in the `model` field with the exact name of the downloaded model (`qwen2.5-coder:7b-instruct-q4_K_M`). Save: Continue.dev reloads the configuration without restarting.",{"title":74,"body":75},"Test the completion from VS Code","Open a code file, place the cursor after a partial function signature and wait for the inline completion. Or open the Continue chat panel and ask a question about your codebase. The response comes from your VPS — no prompt transits through GitHub’s servers.",{"type":35,"title":77,"body":78},"Option 2 — Tabby: dedicated completion server for a team","Tabby is a self-hosted code completion server with approximately 33 800 GitHub stars and an Apache 2.0 licence for the open-source core. It exposes an API compatible with official plugins for VS Code, JetBrains and Neovim, and manages authentication by token: each developer gets an individual key. That is the main difference from Ollama + Continue.dev: Tabby is designed to serve multiple workstations from a single server, with an administration interface and per-developer usage logs.",{"type":55,"title":80,"steps":81},"Deploy Tabby with Docker Compose",[82,85,88,91],{"title":83,"body":84},"Create the Docker Compose file","Define the service with the official `tabbyml\u002Ftabby` image. For CPU-only use, omit the GPU options and choose a compact model as a startup argument: `--model StarCoder2-3B` or `--model Qwen2.5-Coder-1.5B`. Mount a volume to persist downloaded models across restarts and expose port 8080 only on `127.0.0.1`.",{"title":86,"body":87},"Start the server and wait for the model to load","Run `docker compose up -d` then follow the logs: `docker compose logs -f tabby`. The model downloads on the first start. Tabby confirms it is operational when it displays the listening port in the logs.",{"title":89,"body":90},"Configure the reverse proxy and enable authentication","Point a subdomain to `localhost:8080` via Caddy or Nginx Proxy Manager with TLS. In the Tabby administration interface, enable user management, create the administrator account and generate individual access tokens for each developer.",{"title":92,"body":93},"Install the VS Code plugin and enter the token","In VS Code, install the **Tabby** extension from the Marketplace. Enter your server URL (e.g. `https:\u002F\u002Ftabby.your-domain.com`) and the personal token in the extension settings. Inline completion activates immediately across all open files.",{"type":35,"title":95,"body":96},"LiteLLM as a multi-model gateway","If you orchestrate several model sources — Ollama on a VPS for everyday work, an external API for complex tasks — LiteLLM unifies everything behind a single OpenAI-compatible entry point. Continue.dev or any OpenAI-speaking client points to LiteLLM, which routes to the right backend according to priority or budget rules.\n\nYou define spending caps per API key, automatic fallbacks when a provider errors, and model rotation from a central configuration file. The LiteLLM VPS template in the ServOrbit catalogue deploys the proxy and its PostgreSQL database in a single command. This is the right layer when you want to keep the ability to switch to a cloud model without changing your editor configuration.",{"type":98,"body":99},"tip","Never expose Ollama directly on the Internet. Close port `11434` with `ufw deny 11434` and let only the reverse proxy connect to it locally. If several people share the same Ollama server, LiteLLM adds an individual API key management layer on top without modifying the Ollama configuration: each developer has their own LiteLLM token, LiteLLM speaks to Ollama internally.",{"type":35,"title":101,"body":102},"Troubleshooting","**Ollama timeout on the first call.** Check that the container is running (`docker ps`) and that the model is loaded (`docker exec ollama ollama list`). A 7B model can take 30 to 60 seconds to load into memory on the first call after a container restart.\n\n**Model too large for available RAM.** Monitor `docker stats` to track memory usage. Switch to a lighter quantisation (`Q3_K_M` or `Q2_K`) or a more compact model (1.5B or 3B).\n\n**Continue.dev cannot find the API.** Check that `apiBase` in `config.json` points to the HTTPS URL of your reverse proxy and not to `localhost:11434`. Test with `curl -I https:\u002F\u002Follama.your-domain.com` to confirm the TLS certificate is valid and the server responds.\n\n**High latency on CPU only.** On a 7B model in Q4 without GPU, expect 5 to 15 tokens per second depending on vCPU count. Reduce the context sent to Continue.dev by limiting the number of open files included in the index (the `contextLength` parameter).",{"type":35,"title":104,"body":105},"What this changes in practice","The shift to AI Credits made billing variable at the moment agentic sessions consume the most. A VPS running Ollama answers with a fixed monthly cost, full control over the models used, and prompts that stay on your infrastructure.\n\nContinue.dev covers the individual case in five minutes without changing IDE. Tabby covers the team case with per-token authentication and centralised administration. LiteLLM slots in if you orchestrate several model sources from a single point. These three tools are complementary, not competing: the choice depends on the number of developers, the need for auditing and the diversity of models to orchestrate.","Inference engine ready on VPS, at a fixed cost","The Ollama template in the ServOrbit catalogue deploys the inference engine on a Cloud VPS in a single command. OpenAI-compatible API, port exposed locally only, persistent volume for models. Add Continue.dev to VS Code and your coding assistant runs on your own infrastructure.","Activate Ollama on VPS","\u002Fmarketplace\u002Fartificial-intelligence\u002Follama",[111,126,147],{"id":112,"slug":113,"slugs":114,"title":117,"excerpt":118,"readTime":119,"views":15,"isPinned":16,"publishedAt":120,"category":121,"categories":122,"featuredImage":26,"bgImage":27,"posterImage":124,"relatedSolution":125},11,"how-to-host-ollama-on-a-vps",{"fr":115,"en":113,"ar":116},"heberger-ollama-vps","كيفية-استضافة-ollama-على-خادم-vps","Hosting Ollama on a VPS: advanced operational guide","Advanced Ollama VPS setup: model management, nginx reverse proxy, API security, Q4_K_M\u002FQ8 quantization, CVE-2026-45672 Open WebUI and secure coupling.",10,"2026-06-09T00:00:00+00:00",{"id":19,"name":20,"slug":21,"color":22,"icon":23},[123],{"id":19,"name":20,"slug":21,"color":22,"icon":23},"\u002Fblog\u002Fcovers\u002Fheberger-ollama-vps-poster.svg",{"categorySlug":30,"appSlug":31},{"id":127,"slug":128,"slugs":129,"title":132,"excerpt":133,"readTime":134,"views":15,"isPinned":16,"publishedAt":135,"category":136,"categories":142,"featuredImage":26,"bgImage":27,"posterImage":144,"relatedSolution":145},156,"self-host-litellm-on-a-vps-your-private-llm-api-gateway",{"fr":130,"en":128,"ar":131},"self-host-litellm-vps","استضف-litellm-ذاتيا-على-vps-بوابة-api-خاصة-لنماذج-llm","Self-Host LiteLLM on a VPS: Your Private LLM API Gateway","Deploy LiteLLM on a ServOrbit VPS — a self-hosted, OpenAI-compatible gateway to Ollama, Anthropic, Azure and 100+ LLM providers, with spend tracking and rate limits.",4,"2026-01-21T00:00:00+00:00",{"id":137,"name":138,"slug":139,"color":140,"icon":141},7,"Self-hosting","self-hosting","bg-indigo-500\u002F10 text-indigo-400","cloud",[143],{"id":137,"name":138,"slug":139,"color":140,"icon":141},"\u002Fblog\u002Fcovers\u002Fself-host-litellm-vps-poster.svg",{"categorySlug":30,"appSlug":146},"litellm",{"id":148,"slug":149,"slugs":150,"title":153,"excerpt":154,"readTime":155,"views":15,"isPinned":16,"publishedAt":156,"category":157,"categories":158,"featuredImage":26,"bgImage":27,"posterImage":160,"relatedSolution":26},214,"how-to-deploy-an-mcp-server-on-a-vps",{"fr":151,"en":149,"ar":152},"mcp-serveur-ia-auto-heberge-vps","كيفية-نشر-خادم-mcp-على-vps","How to deploy an MCP server on a VPS","The MCP 2026-07-28 spec has made Model Context Protocol the universal standard for AI agents. Here is how to self-host your own MCP server on a VPS.",5,"2026-08-03T00:00:00+00:00",{"id":19,"name":20,"slug":21,"color":22,"icon":23},[159],{"id":19,"name":20,"slug":21,"color":22,"icon":23},"\u002Fblog\u002Fcovers\u002Fmcp-serveur-ia-auto-heberge-vps-poster.svg",1787580997791]