Skip to content
Back to search

PRAXIST

Autonomous research system for measurable, computer-executable research.

Safety scan

A static text scan, not a guarantee. Select the badge to see what was scanned and every finding.

Checked today
Badge for your README
findsafeskills safety rating

Markdown

[![findsafeskills safety rating](/api/listings/sapientinc--praxist/badge.svg)](/listings/sapientinc--praxist)

HTML

<a href="/listings/sapientinc--praxist"><img src="/api/listings/sapientinc--praxist/badge.svg" alt="findsafeskills safety rating" /></a>

Install

i

Install instructions from the README

As written by the repo's author. findsafeskills has not verified these steps.

Install the complete runtime integrations and finish first-use setup with one command:

python3 -m pip install --index-url https://pypi.org/simple "praxist[agents,codex]" && praxist setup --interactive --install-skills codex

The local wizard covers the Fair Source License, User Agreement, privacy, runtime profile, masked credentials, Codex skills, writable examples, and readiness checks. It does not select a research project or launch a run. For Claude Code, use the [host-specific one-line command](docs/getting-started/installation.md#install-and-configure).

For an agent-managed installation, open Codex:

codex --yolo

Then ask it to install and configure Praxist using the packaged OOBE runbook, and to stop after readiness checks.

Before starting research, read the [Quickstart](docs/getting-started/quickstart.md) and [Your First Task](docs/getting-started/first-task.md). They describe the separate takeover step and the project contract it creates.

Choose **Codex-native mode** to use an existing Codex subscription without an API key. For sustained research, Praxist generally favors [open-source model APIs](docs/guides/open-source-model-apis.md) with a high observed cache-hit rate. The setup wizard also supports other API-backed profiles.

What was scanned

The files the safety scan read, as written by the repository's author and shown as plain text. Lines with a finding are marked.

README.md · README · 344 lines · 1 finding marked

View this file on GitHub (read at the repository's current version: the scan did not record which commit it read)

  1. <p align="center">
  2. <img src="docs/assets/brand/praxist-banner.svg" alt="Praxist" width="800">
  3. </p>
  4. <h1 align="left">Praxist: meet your personal R&amp;D team</h1>
  5. <p align="center">
  6. <a href="../../actions/workflows/ci.yml"><img src="../../actions/workflows/ci.yml/badge.svg" alt="CI"></a>
  7. <a href="https://praxist.sapient.inc/en/docs"><img src="https://img.shields.io/badge/docs-open-334155" alt="Documentation"></a>
  8. <a href="https://arxiv.org/abs/2608.25955"><img src="https://img.shields.io/badge/arXiv-2608.25955-B31B1B.svg" alt="arXiv paper"></a>
  9. <a href="https://discord.gg/sapient"><img src="https://img.shields.io/badge/Discord-Join-5865F2?logo=discord&amp;logoColor=white" alt="Discord"></a>
  10. <img src="https://img.shields.io/badge/Python-3.11%2B-3776AB" alt="Python 3.11+">
  11. </p>
  12. Praxist is an autonomous research system for measurable, computer-executable
  13. research. It coordinates parallel research peers, task-owned evaluation,
  14. durable evidence, and generation-to-generation synthesis.
  15. Praxist treats research as a persistent process rather than a sequence of
  16. disconnected prompts. Use it when a project already runs and its objective is
  17. measurable, but the best path forward is still unknown.
  18. <p align="center">
  19. <img src="docs/assets/figures/praxist-overview.svg" alt="Praxist architecture" width="920">
  20. </p>
  21. ## Install Praxist
  22. Install the complete runtime integrations and finish first-use setup with one
  23. command:
  24. ```bash
  25. python3 -m pip install --index-url https://pypi.org/simple "praxist[agents,codex]" && praxist setup --interactive --install-skills codex
  26. ```
  27. The local wizard covers the Fair Source License, User Agreement, privacy,
  28. runtime profile, masked credentials, Codex skills, writable examples,
  29. and readiness checks. It does not select a research project or launch a run.
  30. For Claude Code, use the
  31. [host-specific one-line command](docs/getting-started/installation.md#install-and-configure).
  32. For an agent-managed installation, open Codex:
  33. ```bash
  34. codex --yolo
  35. ```
  36. Then ask it to install and configure Praxist using the packaged OOBE runbook,
  37. and to stop after readiness checks.
  38. Before starting research, read the [Quickstart](docs/getting-started/quickstart.md)
  39. and [Your First Task](docs/getting-started/first-task.md). They describe the
  40. separate takeover step and the project contract it creates.
  41. Choose **Codex-native mode** to use an existing Codex subscription without an
  42. API key. For sustained research, Praxist generally favors
  43. [open-source model APIs](docs/guides/open-source-model-apis.md) with a high
  44. observed cache-hit rate. The setup wizard also supports other API-backed
  45. profiles.
  46. ## Use Praxist Through Codex
  47. We recommend Codex as the interface for operating Praxist. Praxist is not a
  48. replacement for Codex: Codex remains the interactive agent that understands
  49. your project, communicates with you, and uses development tools. Praxist adds
  50. the persistent research loop, parallel peers, evidence protocols, scheduling,
  51. and lifecycle control.
  52. After installation, open Codex in the root of an already runnable research
  53. project and invoke `$praxist-takeover`. The takeover skill inspects readiness,
  54. creates or repairs the task harness, validates its evaluator and evidence
  55. contract, and launches the run after the required gates pass. A precise brief
  56. produces a better research plan; include the objective, metrics, constraints,
  57. resources, exploration choices, and whether launch is authorized.
  58. <details>
  59. <summary>Example takeover brief</summary>
  60. ```text
  61. $praxist-takeover
  62. Treat the current directory as the existing runnable research project. Verify
  63. the baseline and its evaluation path before changing anything.
  64. Optimize <primary metric and direction> while preserving <key constraints>.
  65. Use <peer count> peers for up to <generation count> generations within
  66. <time or cost budget>. Use the runtime and model provider selected during
  67. setup. <Allow or disable> literature search, <enable or disable> QD, and
  68. <enable or disable> generation-zero DIG.
  69. Do not download new datasets or replace required project assets. Build a
  70. separate task harness with explicit metric directions, baseline provenance,
  71. protocol-integrity checks, evidence maturity rules, and justified retention
  72. lanes. After readiness checks pass, <launch immediately in detached mode / ask
  73. for confirmation>. Report the task path, run ID, evidence contract, generation
  74. close policy, and monitor command.
  75. ```
  76. </details>
  77. Other bundled skills:
  78. | Skill | Purpose |
  79. |---|---|
  80. | `praxist-takeover-codex` | No-key takeover using the saved Codex login |
  81. | `praxist-onboarding` | Explain Praxist and inspect local readiness |
  82. | `praxist-task-initialization` | Build or repair a task harness without launching |
  83. | `praxist-interactive-task-init` | Design a task through confirmation-first setup |
  84. | `praxist-control` | Start, stop, resume, monitor, and inspect runs |
  85. | `praxist-diagnostic` | Diagnose run health and produce reports |
  86. | `praxist-scientific-research` | Gather sourced literature and benchmark context |
  87. | `praxist-runtime-install` | Install or repair runtime dependencies and credentials |
  88. | `terminal-line-plot` | Draw metric trends in the terminal |
  89. See [Agent Skills](docs/user-guide/skills.md) for invocation syntax and the
  90. generated [Skills Reference](docs/reference/skills.md) for the complete
  91. contracts.
  92. ## What Praxist Provides
  93. | Capability | Purpose |
  94. |---|---|
  95. | Parallel research peers | Explore competing hypotheses and implementations concurrently |
  96. | Multi-generation synthesis | Carry useful evidence and strategy into later generations |
  97. | Durable evidence lanes | Preserve candidates through incubator, frontier, and Gems state |
  98. | Multi-metric evaluation | Rank task-defined evidence, including Pareto-optimal tradeoffs |
  99. | [Quality-Diversity (QD)](docs/guides/qdig-cohort-allocator.md) and optional [Deep Innovation Gate (DIG)](docs/guides/deep-innovation-gate.md) | Maintain diversity without forcing one exploration policy |
  100. | Central resource scheduling | Adapt experiment admission to observed resource pressure |
  101. | Resume, replay, and monitoring | Keep long-running research inspectable and recoverable |
  102. | Plugin boundaries | Support multiple runtimes, providers, tools, budgets, and workflows |
  103. ## Praxist And The Task Project
  104. | Praxist owns | The task project owns |
  105. |---|---|
  106. | Research orchestration, lifecycle, evidence protocols, replay, scheduling, and extension interfaces | Research objective, executable code, evaluator, metrics, baselines, prompts, roles, and domain constraints |
  107. Praxist contains no task-specific scientific assumptions. A task remains the
  108. single source of truth for what should be tested and what counts as valid evidence.
  109. ## Operate A Run
  110. ```bash
  111. praxist status --json
  112. praxist --monitor --latest
  113. praxist stop <run_id>
  114. praxist resume <run_dir>
  115. ```
  116. `Ctrl-C` closes only the monitor; it does not stop the research run.
  117. ## Examples And Templates
  118. ```bash
  119. praxist examples list
  120. praxist examples install rocket_booster_recovery
  121. praxist examples install rocket_booster_recovery_rust
  122. ```
  123. Complete examples are writable reference projects. `templates/tasks/` contains
  124. replaceable scaffolding for building new task harnesses. The two Rocket Booster
  125. Recovery examples demonstrate the same research problem through Python/JAX and
  126. native Rust implementations.
  127. ## Requirements
  128. | Status | Requirement |
  129. |---|---|
  130. | Required | CPython 3.11+ |
  131. | Required to launch research | A runnable project with measurable evaluation |
  132. | Required for skill-driven operation | Codex or Claude Code; direct CLI operation remains available without either |
  133. | Authentication: choose one | A saved Codex login for Codex-native mode, or a supported provider API key |
  134. | Continuously release-tested | Linux on CPython 3.11 and 3.12 |
  135. | Compatibility target | macOS and other CPython 3.11+ environments; run `praxist doctor` before research |
  136. Task-specific datasets and compute dependencies remain owned by the task project.
  137. See the [platform support matrix](docs/operations/platform-support.md) for the
  138. difference between release-qualified hosts and compatibility targets.
  139. ## Documentation
  140. Read the **[Praxist documentation](https://praxist.sapient.inc/en/docs)**
  141. or open it with:
  142. ```bash
  143. praxist docs
  144. ```
  145. No local documentation server is required.
  146. ## Contributing To Praxist
  147. Read the [contribution guide](.github/CONTRIBUTING.md) together with the
  148. [Code of Conduct](.github/CODE_OF_CONDUCT.md) before participating. Source
  149. maintainers should also follow the architecture and maintenance contract in
  150. [AGENTS.md](AGENTS.md). See the [Privacy Notice](docs/legal/PRIVACY.md) for
  151. Praxist's data-handling terms.
  152. Contact: praxist@sapient.inc
  153. ## FAQ
  154. <details>
  155. <summary>Show questions and answers</summary>
  156. ### Q1. What is Praxist?
  157. Praxist is an autonomous research system for measurable research problems that
  158. can be executed on a computer. It turns an already runnable project into a
  159. continuous, evidence-driven research run.
  160. Across successive generations, parallel research agents develop candidate
  161. solutions; evaluators convert results into structured evidence; and a planning
  162. panel synthesizes that evidence into the research agenda for the next
  163. generation. The cycle continues until the search converges or the budget is
  164. exhausted.
  165. You provide a runnable project and a measurable objective. Praxist orchestrates
  166. the research process that searches for the best-performing solution.
  167. ### Q2. How is Praxist different from manual tuning or AutoML?
  168. AutoML tunes parameters within a predefined search space. Praxist runs the full
  169. research loop.
  170. Parallel research agents can change methods, architectures, and strategies.
  171. Evidence from evaluation shapes the agenda for the next generation, while the
  172. [Deep Innovation Gate (DIG)](docs/guides/deep-innovation-gate.md) and
  173. [Quality-Diversity (QD)](docs/guides/qdig-cohort-allocator.md) allocation help
  174. the system escape local optima.
  175. Praxist is closer to a self-directing research team than a search tool. If your
  176. researchers are already iterating on a problem manually, Praxist takes over the
  177. iteration loop itself.
  178. ### Q3. Is my project a good fit for Praxist?
  179. Praxist delivers the most value when three conditions are met:
  180. - **The objective is measurable:** there is at least one metric that
  181. meaningfully distinguishes better from worse, with a clear optimization
  182. direction.
  183. - **The project already runs:** the baseline code, environment, and required
  184. data or simulator are in place and work without Praxist.
  185. - **The best path forward is unknown.**
  186. If a prerequisite is missing, Praxist stops and tells you exactly what is
  187. needed. It will not silently download unspecified datasets, invent a simulator,
  188. or fabricate baseline performance. That is a deliberate design principle.
  189. ### Q4. Do I need an API key, and what will it cost?
  190. No API key is required in Codex-native mode; Praxist uses your authenticated
  191. Codex session. We also recommend using your own API key to access supported
  192. model APIs.
  193. API costs are set by the provider and vary by model and usage. Total cost also
  194. depends on parallelism, the number of generations, and evaluation runtime. For
  195. cost-sensitive runs, start with a small representative workload before scaling
  196. up.
  197. ### Q5. How does Praxist protect my code and data?
  198. Praxist provides three layers of protection:
  199. - **Project isolation:** Praxist does not modify your original project. Run
  200. artifacts are stored separately.
  201. - **Credentials:** API keys are entered through a masked local prompt and are
  202. not exposed in commands, shell history, or conversations.
  203. - **Data collection:** Praxist does not collect data used in your experiments.
  204. It collects only limited system-level operational information, which you can
  205. disable at any time.
  206. ### Q6. How can I trust that a reported improvement is real?
  207. Praxist uses three safeguards:
  208. - **Preregistration:** Metrics, evaluation protocols, baselines, and acceptance
  209. thresholds are defined before the run.
  210. - **Consistent evaluation:** Every candidate is measured through the same
  211. evaluator, and invalid or suspicious results are excluded.
  212. - **End-to-end provenance:** Every reported improvement includes the evidence
  213. and lineage needed to inspect and reproduce it.
  214. We recommend reviewing what the selected solution changed and testing it again
  215. in your own environment. Praxist's results are designed to be verifiable, and
  216. your own validation should be the final test.
  217. ### Q7. What if Praxist does not improve the result?
  218. Praxist does not guarantee a specific metric improvement. It provides a
  219. rigorous research process and auditable evidence.
  220. If a run does not meet its target, you still receive a negative-result evidence
  221. package, an audit report, and recommendations on whether to stop or redirect
  222. the research.
  223. A negative result can still be valuable: it rules out tested approaches with
  224. evidence and helps prevent further investment in an unproductive direction.
  225. ### Q8. Is Praxist open source, and what terms apply to its outputs?
  226. Praxist is licensed under the Fair Source License Agreement 1.0. The precise
  227. description is source-available: the complete source code is publicly
  228. available and may be viewed, downloaded, and modified. Subject to the license
  229. terms, Praxist may be used for internal business purposes and deployed within
  230. your own organization.
  231. Organizations with aggregate annual revenue, including revenue from
  232. affiliates, below US$1 million may use Praxist commercially at no charge. Once
  233. annual revenue reaches or exceeds that threshold, the organization must
  234. contact the Licensor, Sapient Intelligence Pte Ltd, to negotiate a Commercial
  235. License.
  236. The revenue threshold does not apply to qualifying teaching and academic
  237. research conducted by institutions of higher education, public research
  238. institutions, and nonprofit academic research organizations.
  239. **Generated outputs:** no attribution is required for internal use. If an
  240. output is published externally or otherwise made available to third parties,
  241. the product-name attribution "Praxist by Sapient Intelligence" must be
  242. retained.
  243. This FAQ is a summary only. If it conflicts with the Fair Source License
  244. Agreement 1.0, the terms of the license agreement control.
  245. </details>
  246. ## Citation
  247. If you use Praxist in your research, please cite:
  248. ```bibtex
  249. @misc{li2026praxistexperimentalartifactssolution,
  250. title={Praxist: From Experimental Artifacts to Solution Lineages},
  251. author={Jin Li and Ahmed Murtadha and Zhiyu Wang and Qiwen Chen and William Chen and Yifei Wu and Guan Wang and Andy L. Siy and Jiayi Yang and Mengsha Huang and Wenhao Li and Yixuan Liu and Shuailin Pan and Mingli Yuan and Sen Song and Yuhao Sun},
  252. year={2026},
  253. eprint={2608.25955},
  254. archivePrefix={arXiv},
  255. primaryClass={cs.MA},
  256. url={https://arxiv.org/abs/2608.25955},
  257. }
  258. ```

Ratings and comments

Sign in to rate this skill.

Sign in

Free to use. If it’s helped, pass it on.