Deployment flexible
Instrumentation runs inside your Python agent process, wherever you host it — not tied to a single host or orchestrator.
Developers
Build agents. Emit signals the control plane trusts.
Instrument once. Govern at runtime. PUVINOISE™ brings telemetry and behaviour signals into a multi-tenant control plane — so operators cut MTTR instead of guessing from stderr.
SDK
A thoughtful SDK that gets out of your way
Python-native and framework-agnostic, configured through PUVINOISE_* environment variables. Start agent governance without rewriting your stack.
Python native
pip install puvinoise-sdk, with optional OpenAI and Anthropic extras.
Framework agnostic
LangChain, CrewAI, AutoGen, LangGraph, and more.
Environment configured
bootstrap() reads agent identity and keys from PUVINOISE_* variables.
Deployment flexible
Runs inside your Python agent process, wherever you host it.
# Install the PUVINOISE SDKpip install puvinoise-sdkAdoption without rewrites
Framework and provider agnostic — low friction to govern
Agent stacks change. Vendors change. What must not change is your ability to observe, decide, and recover while agents are live. PUVINOISE™ attaches at runtime so governance starts without a platform migration.
Framework agnostic
LangChain, LangGraph, CrewAI, AutoGen, LlamaIndex, and custom runtimes. Adapters and OTel paths meet you where the agent already runs.
Provider agnostic
Model and cloud vendors stay your choice. PUVINOISE™ does not lock you to one LLM, region, or hosting SKU to get signal, policy, and recovery.
Low-friction onboarding
Install the SDK, emit lifecycle signals, verify in Command Centre. No fleet rewrite, no new agent framework, no months of platform plumbing before governance starts.
Why doing it all at runtime is the northstar
MTTR is the key — and MTTR only moves when detection, policy, and recovery share the live path.
Build-time reviews and static configs cannot see tool failures, policy skips, model drift, or cascading retries as they happen. Runtime is where blast radius grows and where minutes of delay become hours of customer impact. PUVINOISE™ puts instrumentation, behaviour intelligence, governed controls, and recovery on the same runtime path — so operators diagnose with correlated evidence and remediate under policy instead of rebuilding dashboards after every Runtime Case. That is the MTTR northstar: shorten time-to-detect, time-to-decide, and time-to-recover without forcing teams onto a single framework or provider.
Faster time-to-detect: live behaviour signals, not log archaeology
Faster time-to-decide: correlated execution, policy, and proof in one control plane
Faster time-to-recover: governed remediation on the path agents already run
Lower adoption cost: governance starts without a platform migration
Quickstart: instrument, verify, operate
Install and configure
Install the Python SDK (pip install puvinoise-sdk) and set the PUVINOISE_* environment variables for your tenant and agent.
Emit telemetry & signals
Instrument key lifecycle events such as tool invocations, decisions, and errors.
Verify in Command Centre
Confirm signals appear in operational views. Validate correlation IDs and execution context before promoting to broader fleets.
Hand off to operators
Once signals are trusted, SRE and agent ops use behaviour intelligence and Runtime Case workflows — MTTR-minded recovery without bespoke dashboards.
Developer surfaces
Ways to integrate
Meet your fleet where it runs — a Python SDK, a CLI, and OpenTelemetry.
SDK
Instrumentation for agent lifecycles, tool calls, and decisions with minimal boilerplate.
Open SDK docs →CLI
Bind environments, register agents, and run agents with telemetry from the command line.
Programmatic access
Talk to us about programmatic access for your fleet.
OpenTelemetry
Send OpenTelemetry traces — PUVINOISE™ adds behaviour context.
Telemetry reference →Sidecar
An optional sidecar, shipped with the SDK, for binding agent environments.
Framework support
Built-in support for LangChain, LangGraph, CrewAI, AutoGen, and LlamaIndex.
Architecture
From instrumentation to intelligence
Instrumented agents send telemetry to PUVINOISE™, which turns it into behaviour intelligence and governed recovery your operators can act on.
Tenant-scoped
Instrument
Observe
Understand
Govern
Recover
Legend: — tenant boundary · → data flow
Instrumentation layer — not a replacement for your stack
The SDK complements your existing frameworks and deployment model. We document patterns; you retain ownership of agent logic and infrastructure choices — because MTTR improves when governance attaches, not when you migrate.
Open, versioned documentation
Examples that reflect production concerns: tenancy, errors, retries
No requirement to migrate models, frameworks, or hosts to adopt instrumentation
Framework- and provider-neutral by design
A short path from install to Command Centre and governed recovery
Answers
Developer questions, answered directly
Factual answers for engineers instrumenting agents — aligned with the SDK and control plane, not marketing filler.
How do I start agent governance?
Install the PUVINOISE™ Python SDK, emit lifecycle and behaviour signals, and verify ingest in Command Centre. You do not need to rewrite your agent framework or change model providers to start.
Which runtimes, frameworks, and providers does the SDK support?
The SDK is Python. It runs inside your agent process wherever you host it, works with LangChain, LangGraph, CrewAI, AutoGen, and LlamaIndex as well as custom agents, supports OpenAI, Anthropic, and other LLM providers, and exports standard OpenTelemetry.
Why instrument at runtime instead of only at build time?
Build-time reviews cannot see live tool failures or cascading retries. Runtime instrumentation feeds detection, policy, and recovery on the path agents already run — the path that moves MTTR.
Ready to start runtime governance?
Instrument your agents, then talk to developer relations about fleet topology and MTTR outcomes.