Skip to content

Documentation map

Start with the repository overview, then follow the route that matches your decision. You do not need to read every document in this repository; the guides below are separated by the question they answer.

  1. Review the architecture. Read the Architecture overview to understand the component roles, model access paths, and system boundaries.
  2. Deploy a sandboxed harness. Use OpenShell through the manual container guide or bootc deployment guide to deploy OpenCode or OpenClaw on one trusted RHEL host.
  3. Add model routing. Use Praxis through the OpenShell + Praxis guide when the sandboxed harness should use a gateway instead of holding provider credentials.
  4. Choose the inference path. Follow the private vLLM workflow or provider setup after selecting the gateway route.
  5. Reproduce the evidence. Use the testing guide and compatibility matrix before treating a combination as qualified.
  6. Plan the next capability. Use the roadmap to separate accepted behavior from work still in progress.

The separate experimental MaaS + PriceTag pilot documents finite MaaS API keys, USD usage budgets, and PriceTag metering. Its deployment and performance guides are linked from that overview.

Choose by outcome

If you need to… Use
Review component roles and model paths Architecture overview
Install OpenShell and harness containers separately Manual single-server deployment
Choose a deployment method Deployment methods
Combine OpenShell with Praxis OpenShell + Praxis guide
Try content-aware CPU/GPU routing with the optional Web UI llm-d-sc quickstart
Serve users from accounts on one RHEL host All-in-one gateway
Connect harnesses on remote client machines Remote HTTPS/JWT gateway
Try a post-deployment workflow Example scenarios
Inspect or adjust shared model-token allowances Praxis token quotas scenario
Review the MaaS + PriceTag metering pilot Pilot overview, AWS deployment runbook, performance procedure
Deploy the OS image bootc guide
Configure harnesses, providers, or quotas Common configuration index
Inspect OpenShell policies and boundaries OpenShell Overview
Review OpenShell upstream release deltas OpenShell upstream tracking
Reproduce validation or add a new combination Testing guide
Understand future scope and acceptance Roadmap

Choose by component

Component What it provides Start here
OpenShell gateway Policy-controlled execution boundary and sandbox lifecycle for every harness OpenShell Overview
Harness sandboxes Coding agents packaged as digest-pinned images with four profiles: OpenCode, OpenClaw Single-server guide
Praxis Provider credential custody, model routing, and shared token quotas OpenShell + Praxis guide, token quota reference, quota operations scenario
llm-d-sc Experimental semantic prompt classification for CPU/GPU routing llm-d-sc quickstart
MaaS + PriceTag Finite API keys, USD budgets, and usage metering for the experimental single-server pilot Pilot overview, deployment runbook, performance procedure
vLLM Private local model inference on a separate CPU or GPU server Private vLLM workflow
Valkey Durable quota ledger for the managed gateway All-in-one Valkey profile
bootc images Repeatable RHEL 9 OS deployment carrying the selected harness bootc images
Streamlit security demo Credential substitution and filesystem confinement, before and after OpenShell Streamlit demo
OpenCode PR queue Turn public issues or prompts into reviewable patch bundles PR-candidate queue
Operator CLI sss-bootc / sss-praxis and scripts/common/* for install, secrets, quotas, and status bootc operations, all-in-one setup

Harness model access is configured with harness configuration; each harness README lists its profiles and entry points.

Reading rules

For maintainers: build and publish standalone MaaS and PriceTag images.

  • The recommended route is the sandboxed harness → OpenShell → Praxis → inference path. Other guides describe alternate deployment shapes, not interchangeable substitutes.
  • Quickstarts tell you how to run a documented path. Validation records and the compatibility matrix determine what is qualified.
  • Trust-model and quota documents define boundaries that are easy to miss when reading configuration alone.

Upstream references

This repository tracks only its pinned deployment contracts and acceptance evidence. Consult upstream documentation for general project features that this deployment does not qualify.

Next step

Start with the Architecture overview. If you already understand the system and only need a disposable evaluation, use the bootc quickstart.

Preview the documentation locally

From the repository root, install the pinned documentation dependencies and start the local preview:

python3 -m venv .venv-docs
. .venv-docs/bin/activate
python -m pip install -r requirements-docs.txt
scripts/docs/serve

Open http://127.0.0.1:8000/. To refresh the preview after editing docs, run python3 scripts/docs/stage.py; MkDocs rebuilds when .docs-build/ changes. To create a static site in site/, run scripts/docs/serve build instead. The mkdocs.yml navigation includes the PriceTag pilot overview, AWS runbook, and performance procedure.