{"slug": "i-built-a-harness-and-now-it-is-my-daily-driver-for-local-agents", "title": "I built a harness and now it is my daily driver for local agents", "summary": "Developer mrsirg97-rgb released rig, a single-binary local agent runtime that assembles context, streams an OpenAI-compatible model endpoint, executes tool calls, and repeats, shipping 18 built-in tools and a Python plugin system. rig requires Go 1.26.6 or later for the go install path, defaults its endpoint to http://127.0.0.1:8090/v1, and has no model default — a run without one refuses at start. Configuration lives in ~/.rig/ across settings.json, models.json, AGENTS.md, theme.json, and plugins/, with each key resolving in the order flag, environment, file, built-in default.", "body_md": "A minimum runtime for your agents.\n\nOne binary. One model endpoint. One terminal.\n\nrig assembles context, streams the model, executes tool calls, returns results, and repeats. The TUI, piped CLI, headless worker, and dashboard share the same session, task, memory, and scheduler stores.\n\nChoose one:\n\n**Installer** (POSIX sh, no Go, no sudo; installs to `~/.local/bin`):\n\n```\ncurl -fsSL https://mrsirg97-rgb.github.io/rig/install.sh | sh\n```\n\n**Release binary** from `releases/latest`. Choose your `<os>_<arch>`:\n\n```\ncurl -fsSL https://github.com/mrsirg97-rgb/rig/releases/latest/download/rig_linux_amd64 -o rig\nchmod +x rig\n```\n\n**go install** (needs Go ≥ 1.26.6; the core is stdlib-only):\n\n```\ngo install github.com/mrsirg97-rgb/rig/cmd/rig@latest\n```\n\n`rig -update` fetches, verifies, and atomically installs the latest release. The running process keeps the old binary until restart.\n\n```\n./rig --base-url $ENDPOINT --model $NAME\n```\n\nrig needs an OpenAI-compatible SSE endpoint and a model ID. The endpoint defaults to `http://127.0.0.1:8090/v1`; there is no model default — a run without one refuses at start, naming the three ways to set it (`--model`, `RIG_MODEL`, the `model` key in `settings.json`). The TUI is the frontend when stdout is a terminal, the piped CLI otherwise. For scripts, run `./rig -p \"the task\"`. See `docs/SETUP.md` for configuration.\n\nrig ships 18 built-in tools. Restrict them with `--allow`:\n\n| tool | what it does | \n|---|---|\n| `bash` | run shell commands; output bounded | \n| `read` /`write` /`edit` | files; edits are exact-match, provenance-checked | \n| `ls` /`find` /`grep` | the filesystem, by name and by content | \n| `diff` | the working tree against HEAD, or a tool's two latest observations | \n| `python` | a persistent IPython kernel; variables and imports survive | \n| `web_search` | a local SearXNG instance | \n| `web_fetch` | a URL as readable text; private addresses refused | \n| `todo` | the task queue, scoped to the project (a repo's worktrees share one) | \n| `rem` | memory across sessions: learn, recall, reflect, prune; scoped to the project | \n| `scheduler` | background jobs on your crontab, run in a bubblewrap jail | \n| `delegate` | a one-shot headless worker for a bounded subtask | \n| `sessions` | read-only vitals of the session store | \n| `plugin` /`plugins` | the door into your python plugins, and their ecosystem | \n\nEvery tool result is capped. Repeated identical failures are bounded. An optional round cap limits calls per turn. A failed call executes once.\n\nA Python plugin is one file and one tool. It provides `run` and `schema`.\nThere is no build step. Model-authored plugins land in `~/.rig/plugins/pending/`. Approve, disable, and reload them with `/plugins` or the dashboard. See `docs/PLUGINS.md`.\n\nConfiguration lives in `~/.rig/`. Set `$RIG_HOME` to move it. Every file is optional. Invalid files fail startup and name the file and field.\n\n| file | what it holds | \n|---|---|\n| `settings.json` | the knobs: endpoint, model, the allow-list, the retry bound, the approval dial, the worker sandbox | \n| `models.json` | the per-model table: context window, max tokens, the compaction reserve, the role ( `worker` /`interactive` ), the effort levels | \n| `AGENTS.md` | global instructions, read before the project's `<cwd>/AGENTS.md` | \n| `theme.json` | the terminal theme: base, slot colors, glyph set | \n| `plugins/` | your python plugins (top-level files are live) | \n\nEach key resolves in this order: flag, environment, file, built-in default. `/models` lists and switches models. `/effort` changes reasoning effort. See `docs/SETUP.md` for configuration and sandbox settings.\n\n```\nrig serve\n```\n\nThe dashboard serves the rig stores on loopback only. On first run it prints an access token, stores it with mode `0600`, and includes it in the URL. The page exchanges the token for a cookie. Mobile friendly.\n\n- **sessions** : list them per workspace, and resume one mid-work\n- **todo** : the queue, with create, start, complete, and retry\n- **scheduler** : the jobs, with create, pause, resume, remove, an in-place update form that opens with the job's current fields, and each job's run audit trail\n- **models** : the table, with the effort dial\n- **plugins** : approved, pending, disabled; the forge reads and saves a plugin's source into the pending zone\n\n| doc | what it is | \n|---|---|\n| `docs/DESIGN.md` | architecture, the seams, turn semantics, extension guide | \n| `docs/SETUP.md` | build, configuration, verification | \n| `docs/USAGE.md` | running a session; session and failure semantics | \n| `docs/PLUGINS.md` | the python plugins: the contract, the zones, creating and consuming | \n| `SECURITY.md` | the trust model and how to report a vulnerability | \n| `CONTRIBUTING.md` | the process: spec first, tests before code, the freeze | \n\n```\ncmd/rig      composition root; wires every seam once; flags and env only\ncore            the seams, wire types, and the streaming-event vocabulary\nloop            the concrete turn runtime (fault/cancel-aware)\nevt             the event loop (SPEC_EVT): one consumer, many producers; the\n                turn runtime's engine\nkernel.go       the composition kernel\ncommand/        the user commands (/compact, /models, /sessions, /effort, ...)\nconfig/         the four-layer config resolution (flag > env > file > embedded)\nmodels/         the per-model table (window, compaction numbers, role, effort)\npolicy/         ContextPolicy implementations: compact (per-model trigger),\n                effort (the reasoning dial's provider decorator)\nmiddleware/     ToolMiddleware: toolset (the live table), approve (the gate),\n                paths (the ~ boundary), perm (deny by default + plugin\n                provenance), guard (the bound, the round cap, the result cap)\nprovider/       Provider implementations (the openai-compatible SSE adapter)\nplugins/        python plugin discovery (one file, one tool) and the plugin\n                door (run/schema) and the ecosystem (list/create/delete/reload)\nstore/          the SQLite stores (state, todo, rem, scheduler), the sqlx\n                transaction seam, the project scope identity (store/scope);\n                -resume projects a session back from the state rows\ntool/           Tool implementations: bash(1); file read/write/edit; fs\n                ls/find/grep; todo the job queue; rem memory; scheduler\n                background jobs; delegate the one-shot worker; python the\n                persistent IPython kernel; web search and fetch; diff the\n                observation diff; sessions the soak's vitals\nfrontend/       Frontend implementations: cli (the piped reference), tui (the\n                terminal default), oneshot (-p worker), web (the serve\n                dashboard)\nspecs/          the specs, written and agreed before the code (SPEC_CORE first)\ndocs/           DESIGN (architecture), SETUP (build/config), USAGE (running),\n                PLUGINS (the python plugins)\n```\n\nThe structural test is simple: add one file and one registration line. The loop never names a concrete tool, provider, policy, frontend, or middleware. A Python plugin needs no Go. See `docs/DESIGN.md`, `docs/PLUGINS.md`, and `CONTRIBUTING.md` for the process.\n\n`core/` and `loop/` use only the standard library. Stores use the pure-Go `modernc.org/sqlite` driver.", "url": "https://wpnews.pro/news/i-built-a-harness-and-now-it-is-my-daily-driver-for-local-agents", "canonical_source": "https://github.com/mrsirg97-rgb/rig", "published_at": "2026-09-12 20:18:53+00:00", "updated_at": "2026-09-12 20:55:36.575369+00:00", "lang": "en", "topics": ["ai-agents", "ai-tools", "developer-tools", "ai-products"], "entities": ["rig", "mrsirg97-rgb", "Go", "OpenAI", "SearXNG", "IPython", "bubblewrap", "GitHub"], "alternates": {"html": "https://wpnews.pro/news/i-built-a-harness-and-now-it-is-my-daily-driver-for-local-agents", "markdown": "https://wpnews.pro/news/i-built-a-harness-and-now-it-is-my-daily-driver-for-local-agents.md", "text": "https://wpnews.pro/news/i-built-a-harness-and-now-it-is-my-daily-driver-for-local-agents.txt", "jsonld": "https://wpnews.pro/news/i-built-a-harness-and-now-it-is-my-daily-driver-for-local-agents.jsonld"}}