{"id":26119354,"url":"https://github.com/av/boost-starter","last_synced_at":"2026-04-17T21:32:53.661Z","repository":{"id":281544316,"uuid":"945600410","full_name":"av/boost-starter","owner":"av","description":"Minimal starter example for Harbor Boost","archived":false,"fork":false,"pushed_at":"2025-07-06T20:13:58.000Z","size":3739,"stargazers_count":8,"open_issues_count":0,"forks_count":1,"subscribers_count":2,"default_branch":"main","last_synced_at":"2026-04-03T13:52:30.707Z","etag":null,"topics":[],"latest_commit_sha":null,"homepage":null,"language":"Shell","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":null,"status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/av.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":null,"code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null}},"created_at":"2025-03-09T19:46:32.000Z","updated_at":"2025-10-17T17:15:42.000Z","dependencies_parsed_at":"2025-03-09T20:39:20.351Z","dependency_job_id":null,"html_url":"https://github.com/av/boost-starter","commit_stats":null,"previous_names":["av/boost-starter"],"tags_count":0,"template":false,"template_full_name":null,"purl":"pkg:github/av/boost-starter","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/av%2Fboost-starter","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/av%2Fboost-starter/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/av%2Fboost-starter/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/av%2Fboost-starter/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/av","download_url":"https://codeload.github.com/av/boost-starter/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/av%2Fboost-starter/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":31947618,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-04-17T17:29:20.459Z","status":"ssl_error","status_checked_at":"2026-04-17T17:28:47.801Z","response_time":62,"last_error":"SSL_connect returned=1 errno=0 peeraddr=140.82.121.6:443 state=error: unexpected eof while reading","robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":false,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":[],"created_at":"2025-03-10T12:26:18.363Z","updated_at":"2026-04-17T21:32:53.654Z","avatar_url":"https://github.com/av.png","language":"Shell","funding_links":[],"categories":[],"sub_categories":[],"readme":"# boost-starter\n\nThis is a minimal example of getting started with standalone installation of [Harbor Boost](https://github.com/av/harbor/wiki/5.2.-Harbor-Boost) with [custom modules](https://github.com/av/harbor/wiki/5.2.-Harbor-Boost-Custom-Modules).\n\n## One-liners\n\nOne-liners allow launching a Boost instance via Docker with a specific module or configuration.\n\nThis instance is configured to use Ollama running locally on the host on port `11434`, with no API key or `sk-ollama` as the API key. You can adjust that or use any other OpenAI-compatible API endpoint.\n\nAfter starting, you can use Boost's own OpenAI-compatible API:\n```\nAPI_URL=http://localhost:8004/v1\nAPI_KEY=sk-boost\n```\n\nIf your API client runs in Docker, use whatever IP address your Docker host has on the network (most likely `172.17.0.1`).\n\nFor example, in Open WebUI\n\n![boost-webui](./assets/boost-webui.png)\n\n---\n\n# `concept`\n\n![concept](./assets/concept.png)\n\n`concept` is a module allowing LLM to first generate a small concept graph to aid it in replying to the original message.\nThe entire workflow is completely orchestrated so less interesting from interpretability perspective, but more from the representation perspective.\n\n```bash\ndocker run \\\n  -e \"HARBOR_BOOST_OPENAI_URLS=http://172.17.0.1:11434/v1\" \\\n  -e \"HARBOR_BOOST_OPENAI_KEYS=sk-ollama\" \\\n  -e \"HARBOR_BOOST_PUBLIC_URL=http://localhost:8004\" \\\n  -e \"HARBOR_BOOST_MODULES=concept\" \\\n  -p 8004:8000 \\\n  ghcr.io/av/harbor-boost:latest\n```\n\n# `promx`\n\n![promx](./assets/promx.png)\n\n`promx` (Prompt Mixer) implements dynamic metaprompting with real-time control.\n\n\n```bash\ndocker run \\\n  -e \"HARBOR_BOOST_OPENAI_URLS=http://172.17.0.1:11434/v1\" \\\n  -e \"HARBOR_BOOST_OPENAI_KEYS=sk-ollama\" \\\n  -e \"HARBOR_BOOST_PUBLIC_URL=http://localhost:8004\" \\\n  -e \"HARBOR_BOOST_MODULES=promx\" \\\n  -p 8004:8000 \\\n  ghcr.io/av/harbor-boost:latest\n```\n\n# `nbs`\n\n`nbs` (Narrative Beam Search)\n\nImplements a variation of beam search where model generates multiple possible continuations based on a given set of system prompts and then selects the best one.\n\n![nbs](./assets/nbs.png)\n\n```bash\ndocker run \\\n  -e \"HARBOR_BOOST_OPENAI_URLS=http://172.17.0.1:11434/v1\" \\\n  -e \"HARBOR_BOOST_OPENAI_KEYS=sk-ollama\" \\\n  -e \"HARBOR_BOOST_PUBLIC_URL=http://localhost:8004\" \\\n  -e \"HARBOR_BOOST_MODULES=nbs\" \\\n  -p 8004:8000 \\\n  ghcr.io/av/harbor-boost:latest\n```\n\n# `ponder`\n\n`ponder` is similar to the `concept` module above, but with a different approach to building of the concept graph.\n\n![ponder](./assets/ponder.png)\n\n```bash\ndocker run \\\n  -e \"HARBOR_BOOST_OPENAI_URLS=http://172.17.0.1:11434/v1\" \\\n  -e \"HARBOR_BOOST_OPENAI_KEYS=sk-ollama\" \\\n  -e \"HARBOR_BOOST_PUBLIC_URL=http://localhost:8004\" \\\n  -e \"HARBOR_BOOST_MODULES=ponder\" \\\n  -p 8004:8000 \\\n  ghcr.io/av/harbor-boost:latest\n```\n\n# `markov`\n\n![markov](./assets/markov.png)\n\n`markov` renders a completion graph linking tokens in their order of appearance. It produces something similar to a Markov chain for a specific completion and can be used for basic frequency analysis of tokens in the completion.\n\n```bash\ndocker run \\\n  -e \"HARBOR_BOOST_OPENAI_URLS=http://172.17.0.1:11434/v1\" \\\n  -e \"HARBOR_BOOST_OPENAI_KEYS=sk-ollama\" \\\n  -e \"HARBOR_BOOST_PUBLIC_URL=http://localhost:8004\" \\\n  -e \"HARBOR_BOOST_MODULES=markov\" \\\n  -p 8004:8000 \\\n  ghcr.io/av/harbor-boost:latest\n```\n\n# `autotemp`\n\n![autotemp screenshot](./assets/autotemp.png)\n\nThe model will be given a tool to automatically adjust its own temperature based on the specific task.\n\n```bash\ndocker run \\\n  -e \"HARBOR_BOOST_OPENAI_URLS=http://172.17.0.1:11434/v1\" \\\n  -e \"HARBOR_BOOST_OPENAI_KEYS=sk-ollama\" \\\n  -e \"HARBOR_BOOST_MODULES=autotemp\" \\\n  -p 8004:8000 \\\n  ghcr.io/av/harbor-boost:latest\n```\n\n# `dot`\n\n![dot](./assets/dot.png)\n\n`dot` is an extension over \"Draft of Thoughts\" workflow, which makes the LLM to prepare a high-level plan of the response and then execute it iteratively.\n\n# `klmbr`\n\n![klmbr](./assets/klmbr.png)\n\nMakes models more creative (or sometimes just crazy/weird).\nSee the explanation of the approach in the [klmbr litepaper (repo)](https://github.com/av/klmbr)\n\n```bash\ndocker run \\\n  -e \"HARBOR_BOOST_OPENAI_URLS=http://172.17.0.1:11434/v1\" \\\n  -e \"HARBOR_BOOST_OPENAI_KEYS=sk-ollama\" \\\n  -e \"HARBOR_BOOST_MODULES=klmbr\" \\\n  -e \"HARBOR_BOOST_KLMBR_MODS=all\" \\\n  -e \"HARBOR_BOOST_KLMBR_PERCENTAGE=25\" \\\n  -e \"HARBOR_BOOST_PUBLIC_URL=http://localhost:8004\" \\\n  -p 8004:8000 \\\n  ghcr.io/av/harbor-boost:latest\n```\n\n# Plain proxy\n\nWill serve downstream models \"as is\".\n\n```bash\ndocker run \\\n  -e \"HARBOR_BOOST_OPENAI_URLS=http://172.17.0.1:11434/v1\" \\\n  -e \"HARBOR_BOOST_OPENAI_KEYS=sk-ollama\" \\\n  -e \"HARBOR_BOOST_BASE_MODELS=true\" \\\n  -p 8004:8000 \\\n  ghcr.io/av/harbor-boost:latest\n```\n\nYou can combine multiple \"named\" endpoints:\n\n```bash\n# Connects and serves models from Ollama and vLLM\ndocker run \\\n  -e \"HARBOR_BOOST_OPENAI_URL_OLLAMA=https://ollama.foo.com/v1\" \\\n  -e \"HARBOR_BOOST_OPENAI_KEY_OLLAMA=sk-ollama\" \\\n  -e \"HARBOR_BOOST_OPENAI_URL_VLLM=https://vllm.foo.com/v1\" \\\n  -e \"HARBOR_BOOST_OPENAI_KEY_VLLM=sk-vllm\" \\\n  -e \"HARBOR_BOOST_BASE_MODELS=true\" \\\n  -p 8004:8000 \\\n  ghcr.io/av/harbor-boost:latest\n```\n\n\n### Update\n\n`ghcr.io/av/harbor-boost:latest` is updated regularly with new modules and features. Run `docker pull` to update:\n\n```bash\ndocker pull ghcr.io/av/harbor-boost:latest\n```\n\n## Custom Modules\n\nModify the `launch.sh` script to configure your Boost instance. You'll find all supported environment variables documented in the [Boost Wiki](https://github.com/av/harbor/wiki/5.2.-Harbor-Boost#standalone-usage).\n\n\n```bash\n# 1. Clone the repository\ngit clone git@github.com:av/boost-starter.git\n\n# 2. Move to the repository\ncd boost-starter\n\n# 3. Launch boost\n./launch.sh\n```\n\nYou'll find pre-included [`example`](./boost_modules/example.py) module in the `boost_modules` directory with a sample workflow that avoids invoking an LLM altogether and replies with \"Hello, boost!\" to any message.\n\n```python\nID_PREFIX = 'example'\n\nasync def apply(chat, llm):\n  await llm.emit_message('Hello, boost!')\n```\n\nYou can further modify/add files in the `boost_modules` directory to include your custom modules. See the [custom modules guide](https://github.com/av/harbor/wiki/5.2.-Harbor-Boost-Custom-Modules) to learn more.","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fav%2Fboost-starter","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fav%2Fboost-starter","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fav%2Fboost-starter/lists"}