{"slug": "ramabana-rama-s-arrow", "title": "Ramabana – Rama's Arrow", "summary": "Ramabana, a new coding agent from developer Vedic Reader, is now available via pip, featuring a host that coordinates policy, tools, memory, and routing, with models loaded through the rishi tool supporting LiteRT, MLX, llama.cpp, Cursor, and hosted APIs. The agent ships with a terminal app and an MCP server, and its modules are presented as Jupyter notebooks for direct import. Installation is via 'pip install ramabana[cli]', with CPU-only users setting RAMABANA_LITERT_BACKEND=cpu.", "body_md": "The brain of a coding agent: policy, tools, memory, and routing, with no editor attached.\nHosts speak one protocol. Models arrive through [rishi](https://github.com/vedicreader/rishi)\n(LiteRT, MLX, llama.cpp, Cursor, hosted APIs). A terminal app and an MCP server ship as frontends\non that same host.\n\n```\npip install 'ramabana[cli]'\nramabana --root . --model gemma-e4b\nramabana --root . --model gemma-e2b --approve auto 'Reply with exactly: pong'\n```\n\nrishi fetches a model on first use. On CPU-only machines set `RAMABANA_LITERT_BACKEND=cpu`\n\n.\n\n[ LocalHost](https://vedicreader.github.io/ramabana/tools.html#localhost) indexes the folders you open. The harness never offers a tool the host cannot perform.\n\n``` python\nfrom ramabana import Agent\nfrom ramabana.tools import LocalHost\n\nagent = Agent(LocalHost(['..'], web=True), extensions=False)\nagent.ready, agent.note\n(False, 'not started')\npython\nfrom ramabana.testing import fake_agent\na, _ = fake_agent(replies=['in `ramabana/runtime.py`'])\na.ask('where is the compaction threshold?')\n'in `ramabana/runtime.py`'\n```\n\nOne notebook per module: the page you read is the module you import.\n\n[core](/vedicreader/ramabana/blob/main/00_core.ipynb) · [runtime](/vedicreader/ramabana/blob/main/01_runtime.ipynb) · [tools](/vedicreader/ramabana/blob/main/02_tools.ipynb) ·\n[agent](/vedicreader/ramabana/blob/main/03_agent.ipynb) · [cli](/vedicreader/ramabana/blob/main/05_cli.ipynb) · [mcp](/vedicreader/ramabana/blob/main/06_mcp.ipynb) ·\n[vault](/vedicreader/ramabana/blob/main/07_vault.ipynb) · [pyrepl](/vedicreader/ramabana/blob/main/11_pyrepl.ipynb) ·\n[docs](https://vedicreader.github.io/ramabana/)\n\n```\nuv sync --all-extras --group dev\nuv run nbdev-export && uv run pytest\n```\n\n", "url": "https://wpnews.pro/news/ramabana-rama-s-arrow", "canonical_source": "https://github.com/vedicreader/ramabana", "published_at": "2026-08-15 01:03:04+00:00", "updated_at": "2026-08-15 01:11:11.456020+00:00", "lang": "en", "topics": ["ai-agents", "developer-tools", "artificial-intelligence"], "entities": ["Ramabana", "Vedic Reader", "rishi", "LiteRT", "MLX", "llama.cpp", "Cursor", "MCP"], "alternates": {"html": "https://wpnews.pro/news/ramabana-rama-s-arrow", "markdown": "https://wpnews.pro/news/ramabana-rama-s-arrow.md", "text": "https://wpnews.pro/news/ramabana-rama-s-arrow.txt", "jsonld": "https://wpnews.pro/news/ramabana-rama-s-arrow.jsonld"}}