# Get started The open-source toolkit is the starting point. Packs are not for sale. ``` pip install readyagentsdev readyagents new f --from-example calc_pipeline readyagents run f/workflow.yaml ``` Python 3.11–3.14 on Linux, macOS, or Windows. No API keys needed for that first run. Current **2.0.12**. Or clone and `pip install -e .`, then `readyagents run examples/calc_pipeline.yaml`. The wheel ships `examples/`. Current core: [readyagentsdev 2.0.12](https://github.com/readyagentsdev/readyagents-core/releases/tag/v2.0.12). Packs waitlisted, not for sale. Install from the [MCP Registry](https://registry.modelcontextprotocol.io/v0.1/servers?search=io.github.readyagentsdev/readyagents) as `io.github.readyagentsdev/readyagents`. Looking for a config block to paste? [MCP paste catalog](https://github.com/readyagentsdev/readyagents-core/blob/main/docs/mcp-plugs.md) — client plugs only, not a hosted catalog. Typed decisions (experimental, 2.0.12): `readyagents new d --from-example decide_triage` then `readyagents run d/workflow.yaml` (keyless shim heuristic; uncalibrated — low confidence → human gate, exit 2). Resume with `readyagents resume --approve human_review`. Not a general LLM node. `shim` ≠ `jev`. Tried it? Open an [I-ran-this issue](https://github.com/readyagentsdev/readyagents-core/issues/new?template=i-ran-this.md). We are not launching. - [View on GitHub](https://github.com/readyagentsdev/readyagents-core) - [Join the waitlist](/waitlist) ## 01. Install from PyPI pip install readyagentsdev (Current 2.0.12, Python 3.11–3.14) ## 02. Materialize the keyless example readyagents new f --from-example calc_pipeline ## 03. Run it readyagents run f/workflow.yaml — no API keys needed. ## 04. Or clone instead git clone https://github.com/readyagentsdev/readyagents-core.git && cd readyagents-core && python -m venv .venv && source .venv/bin/activate && pip install -e . && readyagents run examples/calc_pipeline.yaml ## 05. Add keys only if you want LLM examples Copy .env.example to .env and set your own provider key. The first run above works without this. ## What’s in the core Local one-shot runs. Not always-on. These are in readyagentsdev 2.0.12: local YAML/JSON workflows with persist/resume (BYOK), the MCP toolkit over stdio by default, opt-in loopback Streamable HTTP with MCP task handles, localhost browser approval, JSON run records with `readyagents runs migrate` to SQLite, and sandboxed builtins. Package status is Beta. The major version covers the core contract only; every extra carries its own tier (stable / preview / experimental) in [`docs/stability.md`](https://github.com/readyagentsdev/readyagents-core/blob/main/docs/stability.md). - Approval (HITL) — a run can pause for approve or reject, then resume. Exit **2** means a human gate is waiting, not that it crashed. Keyless pause: `readyagents new demo --from-example approval_gate` then `readyagents run demo/workflow.yaml`; continue with `readyagents resume --approve gate` / `--reject gate`. - `type: decide` (experimental) — typed `noul` / `choice` / `score` questions, confidence-gated routing to an approval gate, sealed cassette replay. `shim` ≠ `jev`. Different from the CLI verb `readyagents decide`. Not a general LLM node. - `--sovereign` preflight refuses hosted `jev` and names the node; policy `deciders.` allow/deny with `allow_models` / `on_tainted` - Persist + resume — state is written after each node; continue from the last success - Fan-out — `parallel` branches - Include — run another workflow file - `foreach` — bounded sequential loop over a list - Agent `tools:` — allowlisted tool-use loop - Include / parallel resume — successful children are not re-run - MCP toolkit — `readyagents mcp serve` defaults to stdio - Opt-in loopback Streamable HTTP + MCP task handles (start/poll/decide/cancel) - Localhost browser approval — signed decisions, same audit path as the CLI - JSON run records + `readyagents runs list` / `show` / `report` / `replay` / `fork` / `diff` / `freeze` / `migrate` to SQLite - `readyagents doctor` — prints the resolved runs directory and which source set it (`default` / `READYAGENTS_HOME` / config file) - `readyagents new --list-examples`; `readyagents new my-flow [--template basic|approval|research|pipeline|review|foreach|agent-tools|gated]` - `readyagents validate PATH` — schema-validate with source-located errors; catches `decide` question-shape errors offline - `readyagents import` n8n / LangGraph / CrewAI / trigger-action — structural translation only, not behavioural equivalence - Agent firewall (preview) — taint, tool policy, MCP pinning — defence in depth, not a solution to prompt injection. See [`docs/policy.md`](https://github.com/readyagentsdev/readyagents-core/blob/main/docs/policy.md) and [`docs/security-model.md`](https://github.com/readyagentsdev/readyagents-core/blob/main/docs/security-model.md). - Per-node token/cost tracking and budget limits; `--estimate` / `--max-spend` / `--max-tokens`; `readyagents spend` aggregates. Informational — the provider invoice is the truth. - LLM extras: `pip install "readyagentsdev[openai]"` | `[anthropic]` | `[mcp]` | `[all]`. Agent nodes need a key in `.env` after you install the matching extra. - `--dry-run` does not write files - File tools default to the workflow directory (`READYAGENTS_WORKSPACE` still wins) - Successful runs print `run_id:` - Structured JSON logs — `--log-format json` - External decisions — `readyagents decide` / `--decision-file` (no always-on listener) - Append-only audit trail under `$READYAGENTS_HOME/audit/` - `list_dir` — sandboxed directory listing - `readyagents eval` — local fixture suites, no network Try `examples/approval_gate.yaml`, `examples/decide_triage.yaml`, `examples/foreach_calc.yaml`, `examples/agent_tools.yaml`, `examples/multi_gate.yaml`, `examples/eval/pass.yaml`, `examples/list_dir.yaml`, and `examples/connector_demo.yaml`. `--json` prints machine-readable output on `run`, `resume`, `validate`, `eval`, `packs`, and `runs show`. `--log-level` is `DEBUG` / `INFO` / `WARNING` / `ERROR`. Install with `pip install readyagentsdev`, or clone and `pip install -e .`. The wheel ships `examples/`; pip-only first run can use `readyagents new NAME --from-example` or `readyagents new`. ## [Why ReadyAgents?](/why-readyagents) Where MCP connectivity stops and durable orchestration begins, with honest verdicts for ReadyAgents 2.0.12. ## Extras 40+ opt-in extras ship in the same wheel (teams, studio, browser use, knowledge, distillation, registry, environments, A2A, connectors, …). Each has a tier and a limits line in [`docs/extras.md`](https://github.com/readyagentsdev/readyagents-core/blob/main/docs/extras.md). None is required; none changes the core path. Not a marketplace; not a hosted product. ## Bring your own API keys ReadyAgents is designed so you keep provider accounts and spend. Core-only BYOK. Packs waitlisted, not for sale. Use the key you have. ## Want always-on agents? When a workflow has to stay up, join the waitlist. Packs are not for sale. [Join the waitlist](/waitlist)