{"slug": "show-hn-kdeps-a-new-cli-coding-agent-that-runs-on-open-weight-models", "title": "Show HN: Kdeps – A New CLI Coding Agent That Runs on Open Weight Models", "summary": "Kdeps, a new CLI coding agent that runs on open-weight models and is a member of the NVIDIA Inception program, allows developers to run AI workflows locally or deploy them as Docker, Kubernetes, or a single binary without an API key when using Ollama or llamafile. The tool supports multiple backends, multi-agent orchestration, and deterministic DAG pipelines defined in YAML, with features including error handling, resource wiring, and a registry for reusable packages.", "body_md": "## Local AI agent\n\nRun `kdeps` and you are in an AI REPL. Use Ollama or llamafile for a fully offline, private coding agent - no API key, no cloud dependency.\n\nRun AI workflows locally. Or deploy them anywhere.\n\nInstall kdeps, run `kdeps`, get an AI agent - no API key needed with Ollama or llamafile. Build your workflow in YAML. Deploy as Docker, Kubernetes, or a single binary when you're ready. Proud member of the NVIDIA Inception program.\n\n``` bash\n$ kdeps\n\nkdeps v2.x.x  |  Local agent mode\nModel: llama3.2 (Ollama)  |  Type /help for commands\n\n> write a Go function that parses a CSV file\n\nSure. Here's an idiomatic Go CSV parser...\n\nfunc ParseCSV(r io.Reader) ([][]string, error) {\n    reader := csv.NewReader(r)\n    return reader.ReadAll()\n}\n\n> /model claude-opus-4-8\nSwitched to claude-opus-4-8 (Anthropic)\n\n>\n```\n\ngetting started\n\nThree steps from idea to running AI API.\n\nDeclare resources -- chat, HTTP, Python, SQL, shell. Wire them with `requires:`\n\n. No glue code.\n\n`kdeps run workflow.yaml`\n\nstarts the API server. `kdeps serve`\n\nstarts the autonomous agent loop.\n\nPOST to your route, get back structured JSON. Export as Docker, Kubernetes, ISO, or a single binary.\n\nplatform agnostic\n\nSwitch backends in config. No code changes, no lock-in.\n\nrun modes\n\nWorkflows, agents, and agencies — all from the same YAML.\n\nResources run in DAG order defined by `requires:`\n\n. Every request takes the same path. Predictable, auditable, ships to production.\n\nRun `kdeps`\n\nfor an instant AI REPL - fully offline with llamafile or Ollama, no API key needed. Load workflows as tools: the LLM decides what to call and when.\n\nOne agent calls another via the `agent:`\n\nresource type. Compose agents like functions — each runs independently, results flow back.\n\nwhy kdeps\n\nNo Python scripts, no wiring, no boilerplate.\n\n| Traditional approach | kdeps |\n|---|---|\n| Python script + Flask + OpenAI SDK + retry logic | One `workflow.yaml` file |\n| Manual dependency wiring | `requires:` in YAML |\n| Write error handling by hand | `onError:` block (continue / retry / fail) |\n| Write Dockerfile + CI pipeline for deployment | `kdeps bundle build --tag` then `docker push` |\n| Glue code between services | Resources pass data via `output()` |\n| Manual polling loop for bots | `input.sources: [bot]` in workflow config |\n| Chaining agents by hand | `agent:` resource — one agent calls another declaratively |\n\nregistry\n\nInstall pre-built packages from the registry. Publish your own.\n\nReusable capability extensions. Install with one command, invoke with `component:`\n\nand typed inputs.\n\nComplete DAG pipelines packaged as `.kdeps`\n\narchives. Drop them into an agency or run them standalone.\n\nMulti-agent orchestration bundles packaged as `.kagency`\n\narchives. One entry point, many agents.\n\nexamples\n\nReal patterns from the examples directory. Every one is a working workflow.\n\nPOST a JSON body, run a DAG pipeline, get structured JSON back. The default pattern for workflow mode.\n\nworkflow modePoll for messages, run a multi-step pipeline, reply with LLM responses. Two resources: llm and reply.\n\nworkflow modeRead from stdin, call an LLM, write to stdout. One-shot. Perfect for cron jobs and CI pipelines.\n\nworkflow modeIndex documents locally, search with keywords, feed results into an LLM prompt. Fully on-prem.\n\nworkflow modeOrchestrate multiple agents. One summarises, another translates — each an independent workflow.\n\nmulti-agentbook\n\nEverything from first agent to production deployment, in one place.\n\nBuild & Deploy Autonomous AI Agents and Agencies in YAML\n\nYour AI prototype works. Now ship it. Hands-on guide to deterministic pipelines, multi-agent orchestration, error handling, and vendor-agnostic deployment — the production challenges most AI frameworks leave to you.\n\nRead the book ->", "url": "https://wpnews.pro/news/show-hn-kdeps-a-new-cli-coding-agent-that-runs-on-open-weight-models", "canonical_source": "https://kdeps.com", "published_at": "2026-07-28 06:11:14+00:00", "updated_at": "2026-07-28 06:22:24.799495+00:00", "lang": "en", "topics": ["ai-tools", "developer-tools", "ai-agents", "ai-infrastructure"], "entities": ["Kdeps", "NVIDIA Inception", "Ollama", "llamafile", "Anthropic"], "alternates": {"html": "https://wpnews.pro/news/show-hn-kdeps-a-new-cli-coding-agent-that-runs-on-open-weight-models", "markdown": "https://wpnews.pro/news/show-hn-kdeps-a-new-cli-coding-agent-that-runs-on-open-weight-models.md", "text": "https://wpnews.pro/news/show-hn-kdeps-a-new-cli-coding-agent-that-runs-on-open-weight-models.txt", "jsonld": "https://wpnews.pro/news/show-hn-kdeps-a-new-cli-coding-agent-that-runs-on-open-weight-models.jsonld"}}