{"slug": "qlaude-claude-code-against-your-own-model-server", "title": "Qlaude – Claude Code against your own model server", "summary": "A new open-source launcher called qlaude lets developers run Anthropic's Claude Code terminal coding agent against their own model server, requiring no Anthropic subscription or account. The POSIX-sh script, installed via a single curl command, stores the endpoint, model and token in ~/.config/qlaude/config.env with 0600 permissions and launches Claude Code against any server that answers Anthropic's Messages API (POST /v1/messages), with vLLM cited as the reference server since v0.11.1. The setup was verified in November 2026 on Linux and macOS with Claude Code 2.1.325 plus vLLM 0.33, and with a llama-swap plus llama.cpp stack serving a Gemma MoE model; Windows is not supported.", "body_md": "Run [Claude Code](https://code.claude.com), Anthropic's terminal coding\nagent, against your own model server. Claude Code is a command-line tool\nthat reads your code, edits files and runs commands. One paste and five\nsetup questions get you running. You need no Anthropic subscription or\naccount, and you do not have to own a GPU or run a server to try it: a\nhosted provider that offers an Anthropic-compatible API endpoint can\nstand in for your own vLLM — see [Prerequisites](https://github.com/trail-openers/qlaude/blob/main/docs/prerequisites.md).\n\nWorks on Linux and macOS. Windows is not supported.\n\nVerified November 2026 on Linux and macOS: Claude Code 2.1.325 + vLLM\n0.33 on the `POST /v1/messages` route, plus a llama-swap + llama.cpp\nstack serving a Gemma MoE — what exactly was tested, and where, is in\n[Status](https://github.com/trail-openers/qlaude/blob/main/docs/status.md).\n\nThe reference server is [vLLM](https://docs.vllm.ai), the open-source\nLLM inference server: since v0.11.1 it answers Anthropic's Messages API\n(`POST /v1/messages`) itself, so Claude Code connects directly, without\na translation server in between — vLLM's\n[official Claude Code integration](https://docs.vllm.ai/en/stable/serving/integrations/claude_code/).\nAnything that speaks the same route works.\n\nYou need three things: a model server that is already running (or a\nhosted provider), its URL, and its access token. This repo does not run\na server for you — if you don't have one yet, read\n[Prerequisites](https://github.com/trail-openers/qlaude/blob/main/docs/prerequisites.md) first.\n\n`qlaude` is the launcher this repo ships: a single POSIX-sh script\n(POSIX sh is the most portable shell dialect). It stores your\nendpoint/model/token in `~/.config/qlaude/config.env` (mode 0600 —\nreadable only by your user), runs the self-checks below as\n`qlaude doctor`, and launches Claude Code with every setting wired for\nyou.\n\n```\ncurl -fsSL https://raw.githubusercontent.com/trail-openers/qlaude/main/install.sh | sh\n```\n\nThat paste installs `~/.local/bin/qlaude` plus\n`~/.config/qlaude/SYSTEM-PROMPT.md` (mode `0600`, the optional behavior\nprompt — see [Behavior prompt](https://github.com/trail-openers/qlaude/blob/main/docs/behavior-prompt.md)). Nothing else:\nno daemon, and no shell-rc edits — the only command that ever touches rc\nfiles is `qlaude ensure-path`, and only with your consent. Every\ndownloaded file is verified against `install.sha256` before running —\ncorruption/drift protection, not first-run authentication (the checksum\nlist arrives over the same HTTPS connection; trust here is TLS plus the\nhost). The token is never echoed, never printed, never a command-line\nargument.\n\nIn a terminal the installer prints the PATH line to copy, if needed, then runs the setup interview. It asks five questions:\n\n1. Endpoint URL — the address of your server; no trailing `/v1` needed,\nit normalizes.\n2. Token — shown as `*` while you type, stored`0600` .\n3. Model — picked from the models the server lists; Enter accepts a single discovered id. If the listing is unavailable, the interview lets you type the id yourself.\n4. Context — how many tokens the model can see at once.\n5. Reasoning effort — how hard the model thinks before answering.\n\nThe interview ends with the five checks below. A successful run looks like this (output sketch — your server URL and model names will differ):\n\n``` bash\n$ curl -fsSL https://raw.githubusercontent.com/trail-openers/qlaude/main/install.sh | sh\nqlaude install: fetching from https://raw.githubusercontent.com/trail-openers/qlaude/main\nqlaude install: checksums verified (truncation/corruption guard; supply chain = TLS + the host)\nqlaude install: installed ~/.local/bin/qlaude\nqlaude install: PATH note: ~/.local/bin is not on PATH — copy-paste now: export PATH=\"$HOME/.local/bin:$PATH\" (then: ~/.local/bin/qlaude ensure-path)\nqlaude install: next: guided configuration — answer the prompts (Ctrl-C skips; resume any time: ~/.local/bin/qlaude setup)\nqlaude: endpoint URL (the Anthropic-compatible server qlaude will route to — type or paste, it echoes back): https://models.example.com\nqlaude: token for https://models.example.com — paste it (shown as *, stored 0600 in config.env, never printed), Enter when done: **********\nqlaude: model — 1 discovered; Enter accepts my-model (or type another id):\nqlaude: context window tokens (= your server's --max-model-len; Enter = 262144):\nqlaude: reasoning effort low/medium/xhigh ('high' is mapped for you; Enter = medium):\nPASS 0 prereqs: all found\nPASS 1 messages: 200 with content text\nPASS 2 claude: echoed OK\nPASS 3 tools: tool call round-tripped\nPASS 4 info: no informational issues\n```\n\nAnd that is the whole setup: you set **no environment variables** by\nhand. The launcher exports everything Claude Code needs at launch; the\nfull reference for each one is\n[Environment variables](https://github.com/trail-openers/qlaude/blob/main/docs/environment.md), which you only need when\ndebugging or running `claude` without `qlaude`.\n\nDaily use: plain `qlaude` opens Claude Code against your endpoint —\njust type what you want done. `qlaude doctor` repeats the five checks\nany time; `qlaude config` shows your saved settings (never the token).\nTo hand arguments straight to Claude Code — a `-p` probe, `--resume`,\nanything after `--`:\n\n```\nqlaude -- -p \"Reply with exactly: OK\" --max-turns 1\n```\n\nSkipped or Ctrl-C'd the interview? Rerun with `~/.local/bin/qlaude setup`.\nIn scripts and CI the installer only installs; configure explicitly,\nfeeding the token over stdin — never as an argument (in CI, from your\nsecret store, not shell history):\n\n```\ncurl -fsSL https://raw.githubusercontent.com/trail-openers/qlaude/main/install.sh | sh\nprintf '%s' \"$YOUR_TOKEN\" | ~/.local/bin/qlaude setup -y --token-stdin \\\n  --endpoint https://models.example.com --model my-model\n```\n\nNo subcommand. Three lines. The last one is macOS only and removes the keychain entry:\n\n```\nrm \"$HOME/.local/bin/qlaude\"\nrm -rf \"$HOME/.config/qlaude\"\nsecurity delete-generic-password -s qlaude   # macOS only\n```\n\nIf you installed with `QLAUDE_INSTALL_DIR`/` QLAUDE_CONFIG_DIR` set, drop\nthose files from wherever they went instead. If you ran\n`qlaude ensure-path`, delete the `# qlaude PATH` … `# end qlaude PATH`\nmarker block from the file it named (`~/.profile` by default,\n`~/.zprofile` for zsh and macOS).\n\nRun `qlaude doctor` first — it names the failing check. Then:\n\n- [Troubleshooting](https://github.com/trail-openers/qlaude/blob/main/docs/troubleshooting.md) — known symptoms, real\ncauses and fixes (including the misleading \"model may not exist\"\nerror, which is almost always a trailing`/v1` ).\n- [Verify, in this order](https://github.com/trail-openers/qlaude/blob/main/docs/verify.md) — two commands that split\nserver problems from environment problems.\n- [Server-side checklist](https://github.com/trail-openers/qlaude/blob/main/docs/server-checklist.md) — the vLLM flags\nthat make tools work (`--tool-call-parser` ,`--served-model-name` ,`--max-model-len` ).\n\n- [Prerequisites](https://github.com/trail-openers/qlaude/blob/main/docs/prerequisites.md) — Claude Code install, your\nown vLLM or a hosted provider, and where the token lives.\n- [How it works](https://github.com/trail-openers/qlaude/blob/main/docs/how-it-works.md) — the Messages route, and the\ntwo details behind almost every failure people hit.\n- [Environment variables](https://github.com/trail-openers/qlaude/blob/main/docs/environment.md) — the canonical\nreference for every variable (the normal path sets none).\n- [Auto mode and the safety classifier](https://github.com/trail-openers/qlaude/blob/main/docs/auto-mode.md) — why the\nlauncher exports`CLAUDE_CODE_AUTO_MODE_SERVER=0` , and when to opt\nback in.\n- [Gateways](https://github.com/trail-openers/qlaude/blob/main/docs/gateways.md) — direct is the default; when a proxy\nearns its keep, and which to avoid.\n- [Behavior prompt](https://github.com/trail-openers/qlaude/blob/main/docs/behavior-prompt.md) — what the optional`SYSTEM-PROMPT.md` rules prevent, and what deliberately is not in it.\n- [Status](https://github.com/trail-openers/qlaude/blob/main/docs/status.md) — exactly what has been verified live.\n- Design and rationale: [the spec](https://github.com/trail-openers/qlaude/blob/main/docs/superpowers/specs/2026-11-16-qlaude-design.md) .\n\n- Questions and bugs: [open an issue](https://github.com/trail-openers/qlaude/issues) —\ninclude what you ran, what came back, your OS, and which commit you\ninstalled from (your`QLAUDE_REF` value; if you left it on`main` , the\ninstall date helps too — the installer fetches single files, so there is\nno local clone to query with`git rev-parse` ).\n- Security issues: please do not open a public issue — report privately\nvia [SECURITY.md](https://github.com/trail-openers/qlaude/blob/main/SECURITY.md) (GitHub's*Report a vulnerability* form).\n- Contributions: [CONTRIBUTING.md](https://github.com/trail-openers/qlaude/blob/main/CONTRIBUTING.md) — the offline test suite,\nthe shellcheck/dash gates, the line budgets (hard per-file size caps CI\nenforces), and the README rules.\n\nLicensed under the [Apache License 2.0](https://github.com/trail-openers/qlaude/blob/main/LICENSE) — use it, fork it, ship it\nin your company.", "url": "https://wpnews.pro/news/qlaude-claude-code-against-your-own-model-server", "canonical_source": "https://github.com/trail-openers/qlaude", "published_at": "2026-11-22 07:20:04+00:00", "updated_at": "2026-11-22 07:41:59.591676+00:00", "lang": "en", "topics": ["ai-tools", "ai-agents", "developer-tools", "large-language-models", "ai-infrastructure"], "entities": ["qlaude", "Claude Code", "Anthropic", "vLLM", "llama.cpp", "llama-swap", "Gemma", "trail-openers"], "also_reported_by": [], "takeaway": "Since vLLM v0.11.1 answers Anthropic's Messages API at POST /v1/messages directly, Claude Code can connect to a self-hosted model server without a translation layer, and qlaude wires the endpoint, token, model, context size and reasoning effort into ~/.config/qlaude/config.env at mode 0600.", "recommendation": null, "alternates": {"html": "https://wpnews.pro/news/qlaude-claude-code-against-your-own-model-server", "markdown": "https://wpnews.pro/news/qlaude-claude-code-against-your-own-model-server.md", "text": "https://wpnews.pro/news/qlaude-claude-code-against-your-own-model-server.txt", "jsonld": "https://wpnews.pro/news/qlaude-claude-code-against-your-own-model-server.jsonld"}}