Package Guides
Package-level setup material for the real packages in this repo: what to install, when to use it, and the shortest working setup shape.
Use SDKs & Adapters to choose the right path. Use this page when you already know the package name and want the real install and setup surface.
Start Here
| If you are installing... | Start at |
|---|---|
primust or primust[auto] | Python runtime |
primust-ai | AI adapter meta-package |
primust-checks | Checks harness |
primust-verify | Verifier |
primust-langgraph, primust-openai-agents, primust-google-adk, or primust-otel | Named adapter packages |
@primust/sdk | JavaScript / TypeScript |
primust-connectors | Connectors |
primust-cedar, primust-drools, primust-odm, or primust-opa | Policy and decision engines |
Core Package Map
| Package | Ecosystem | Use it for |
|---|---|---|
primust | Python / PyPI | Core runtime SDK and CLI |
primust-ai | Python / PyPI | Meta-package for AI adapter extras |
primust-checks | Python / PyPI | Open-source checks harness |
primust-verify | Python / PyPI | Local and offline verification |
primust-langgraph | Python / PyPI | LangGraph adapter |
primust-openai-agents | Python / PyPI | OpenAI Agents SDK adapter |
primust-google-adk | Python / PyPI | Google ADK adapter |
primust-otel | Python / PyPI | OpenTelemetry span bridge |
@primust/sdk | npm | JavaScript / TypeScript SDK |
primust-connectors | Python / PyPI | Vendor-boundary regulated platform connectors |
Python Runtime: primust and primust[auto]
This is the shortest path from an existing Python app to a real emitted VPEC.
pip install "primust[auto]" primust-verify
export PRIMUST_API_KEY=pk_live_xxx
python my_app.py
Set the key in your environment or secret manager first. Do not hardcode live keys in source or normalize inline one-off command usage as the recommended path.
- Best for: existing Python apps, agents, and workflows
- Command surface: the package installs the
primustCLI - Optional extras on the core package:
langgraph,openai-agents,google-adk,semantic-kernel,otel,anthropic,openai,sklearn,statsmodels,httpx,requests,litellm,bedrock,torch,transformers, andauto - What you should expect: runtime capture, emitted artifact, and
primust.lock
primust-ai
This is the AI adapter meta-package. Use it when you want a single dependency that can bring in multiple named agent-framework adapters.
pip install "primust-ai[all]"
Available extras are:
[langgraph]
[openai-agents]
[google-adk]
[all]
primust-checks
Open-source checks harness for local preview checks, bundle-based checks, and bring-your-own-check workflows.
pip install primust-checks
Pythonfrom primust_checks import Harness
harness = Harness(policy="ai_agent_general_v1")
result = harness.run(input="Summarize this quarterly report.")
print(result.passed, result.gaps)
- Best for: lightweight starts, local enforcement, custom checks
- Good starter bundles:
ai_agent_general_v1,eu_ai_act_art12_v1,hipaa_safeguards_v1,soc2_cc_v1,coding_agent_v1
primust-verify
Canonical local verifier for VPECs and Evidence Packs.
pip install primust-verify
primust-verify vpec.json --trust-root key.pem
Pythonfrom primust_verify import verify
result = verify(vpec_json)
assert result.valid
- Best for: local review, CI, audit workflows, long-term verification
- Key flags:
--production,--skip-network,--trust-root,--json
Named Python Adapter Packages
primust-langgraph
pip install primust-langgraph
Pythonimport primust
from primust_langgraph import PrimustLangGraph
p = primust.Pipeline(workflow_id="my-agent")
adapter = PrimustLangGraph(pipeline=p)
instrumented = adapter.instrument(compiled_graph)
result = instrumented.invoke({"input": "..."})
Best when LangGraph node boundaries are the governance-relevant execution surface.
primust-openai-agents
pip install primust-openai-agents
Pythonimport primust
from primust_openai_agents import PrimustOpenAIAgents
p = primust.Pipeline(workflow_id="my-agent")
adapter = PrimustOpenAIAgents(pipeline=p)
instrumented_agent = adapter.instrument(agent)
result = instrumented_agent.run("...")
Best when tool calls and handoffs inside the OpenAI Agents SDK are the main evidence surface.
primust-google-adk
pip install primust-google-adk
Pythonimport primust
from primust_google_adk import PrimustGoogleADK
p = primust.Pipeline(workflow_id="my-agent")
adapter = PrimustGoogleADK(pipeline=p)
instrumented = adapter.instrument(agent)
result = instrumented.run("...")
Best when Google ADK tool and callback boundaries are the main governed surface.
primust-otel
pip install primust-otel
Pythonimport primust
from primust_otel import PrimustSpanProcessor
from opentelemetry import trace
from opentelemetry.sdk.trace import TracerProvider
p = primust.Pipeline(workflow_id="my-service")
provider = TracerProvider()
provider.add_span_processor(PrimustSpanProcessor(pipeline=p))
trace.set_tracer_provider(provider)
Best when you already have a real OpenTelemetry tracing surface and want to turn it into evidence rather than duplicate instrumentation.
@primust/sdk
JavaScript and TypeScript SDK with direct exports for pipeline work and browser/server-side verification helpers.
npm install @primust/sdk
- Main export:
@primust/sdk - Verifier export:
@primust/sdk/verify - Key surfaces:
Pipeline,Run, upstream helpers,verifyVpec()
primust-connectors
Python package for regulated vendor-boundary connectors.
pip install primust-connectors
Current connector inventory and real per-platform notes live on the dedicated Connectors page.
Policy and Decision Engine Packages
Cedar, Drools, IBM ODM, and OPA now have a dedicated setup page because they are not edge-case adapters.
primust-cedar
primust-drools
primust-odm
primust-opa
Use Policy & Decision Engines for the real setup flow and engine-specific notes.
Need the path chooser? Go back to SDKs & Adapters. Need the full app flow after install? Read the User Guide. Need the exact field and endpoint model? Use the Technical Reference.