01
Before you install PRAXIST
PRAXIST is not a general chatbot for an empty folder. Sapient’s documentation says it needs an already runnable project, a measurable objective and an environment that can evaluate whether one candidate is better than another.
The public repository currently lists CPython 3.11+ as required. Linux on Python 3.11 and 3.12 is continuously release-tested; macOS and other Python 3.11+ environments are compatibility targets.
02
PRAXIST and Codex are different layers
Sapient recommends Codex as one interface for operating PRAXIST. Codex remains the interactive coding agent that understands the project and communicates with you; PRAXIST supplies the persistent research loop, parallel peers, evidence protocols and lifecycle control.
Claude Code can also be used as the host interface, and direct CLI operation remains available without either host agent.
03
Install PRAXIST for Codex
Sapient’s repository currently publishes this Codex-oriented setup command: python3 -m pip install --index-url https://pypi.org/simple "praxist[agents,codex]" && praxist setup --interactive --install-skills codex.
The setup wizard covers the Fair Source license, user agreement, privacy, runtime profile, credentials, Codex skills, examples and readiness checks. It does not choose a research project or launch a run.
04
Can PRAXIST use your Codex login without an API key?
Yes, according to Sapient’s current repository. Codex-native mode can use an existing authenticated Codex session, so a separate model API key is not mandatory for that path.
API-backed profiles are also supported. Their cost depends on the provider, model, parallel peers, generations and evaluation runtime rather than on one fixed PRAXIST usage price.
05
Using Claude Code
Sapient documents Claude Code as another supported host interface. Because host-specific beta commands can change quickly, use the current repository or official documentation for the exact Claude Code installation command at the time you install.
The architecture remains the same: Claude Code is the interactive agent layer, while PRAXIST supplies the persistent autonomous research process.
06
Run the readiness checks first
On compatibility-target systems, Sapient explicitly recommends running praxist doctor before research. A successful package installation does not mean the research task itself is ready.
The baseline project, data and evaluator should still run correctly outside PRAXIST before you hand the task over.
07
Prepare a good first task
A useful first task needs a metric with a clear optimization direction, such as lower error, higher accuracy, faster runtime, lower cost or a simulation success rate.
- A reproducible runnable baseline
- The command or harness that evaluates a candidate
- The metric and optimization direction
- Constraints that cannot be violated
- The files or components agents may change
- A reasonable compute or generation budget
08
Take over the project from Codex
After setup, Sapient says to open Codex in the root of an already runnable research project and invoke the packaged PRAXIST takeover skill. The takeover step creates the project contract used by the research process.
Installation and takeover are separate: setup prepares the runtime; takeover prepares a particular project for autonomous research.
09
Monitor, stop and resume runs
The repository exposes lifecycle commands including praxist status --json, praxist --monitor --latest, praxist stop <run_id> and praxist resume <run_dir>.
10
Start small before spending heavily
Persistent multi-agent research can use far more compute than a normal coding session. Sapient recommends beginning with a small representative workload before scaling.
11
Check the license before commercial use
PRAXIST is source-available under Sapient’s Fair Source License Agreement 1.0. The current repository says organizations below US$1 million in aggregate annual revenue may use it commercially without charge under the stated conditions, while larger organizations must contact Sapient for a commercial license.
Read the current license itself before business deployment because the license text, not a tutorial summary, controls your rights.
12
Keep the broader product questions on the main page
For broader product status, benchmark caveats, use cases and limitations, return to the main PRAXIST overview. This guide stays focused on installation and first-run workflow.
Sources
Primary and supporting sources
Facts were rechecked against the linked sources immediately before publication. Pricing, product availability and rollout status can change.