Technology · Agentic → Guardrails · Open Source · wiki:deep
NeMo Guardrails
NVIDIA NeMo Guardrails is an open-source toolkit for adding programmable guardrails (“rails”) to LLM conversational apps: control topical boundaries, dialog paths, language style, structured extraction, tool/service connections, and defenses against jailbreaks/prompt injection. Rails sit between application code and the LLM. Install with pip install nemoguardrails (Python 3.10–3.13). Documented use cases include RAG Q&A moderation, domain chatbots, LLM endpoints, and optional LangChain wrapping.
Introduced in the paper NeMo Guardrails: A Toolkit for Controllable and Safe LLM Applications with Programmable Guardrails (arXiv:2310.10501).
Why it matters here
Policy for an agentic OS is not only “block bad strings” — it is declared conversational and tool-use behavior. NeMo Guardrails is the research default for Colang-style programmable rails and vulnerability-scanning evals around chat bots. Prefer Guardrails AI for Hub validator composition; prefer Headroom when measuring HITL escalation for coding-agent actions.
How it works
Developers author a guardrails configuration (flows/rails). At runtime the library intercepts messages, applies rails (including topical and security rails), and may call tools under constrained dialog paths before/after the LLM.
- Install
nemoguardrailsand author a rails config. - Place the rails layer between app and LLM (or around a LangChain chain).
- Runtime enforces paths, topics, and safety rails.
- Evaluate with included vulnerability-scanning tooling where relevant.
Related: guardrails-ai · headroom · llm-guard · topics/13-governance-policy
Scroll inside the canvas to pan
When to reach for it
- Use when: you need programmable conversational/tool rails (Colang configs, topical control, dialog paths).
- Skip when: you only need lightweight Hub validators (guardrails-ai), archived scanners (llm-guard), or action-approval measurement (headroom).
- Prefer instead: Guardrails AI for validator packs; Headroom for coding-agent HITL budgets.
Limits
- Config complexity: rails are powerful and easy to under-test — treat configs as reviewed policy.
- Chat-centric: less natural for non-dialog batch agents without adaptation.
- Not a sandbox: rails ≠ isolated code execution.
- Not the ledger: rail configs support policy; they do not replace declared company authority.
Linked from
Catalog backlinks — what points here (wiki “what links here”).
Features
- access-vs-authority
- autonomy-levels
- emergency-control
- human-oversight-guardrails
- job-healthcare
- job-security
- r2-13-governance-policy
- r2-16-approval-calibration
- security-identity-governance
- wt-human-policy
Stacks
Peers
Document management
What we checked
Claims below are backed by science sources on disk.
Toolkit / programmable rails
NVIDIA-NeMo/Guardrails · MODERATE
“NeMo Guardrails is an open-source toolkit for easily adding programmable guardrails to LLM-based conversational systems.”
Colang / config surface
Colang programmable rails (same paper) · MODERATE
“Differently, using a runtime inspired from dialogue management, NeMo Guardrails allows developers to add programmable rails to LLM applications - these are user-defined, independent of the underlying LLM, and interpretable.”
Research paper
NeMo Guardrails: A Toolkit for Controllable and Safe LLM Applications with Programmable Rails · STRONG
“NeMo Guardrails is an open-source toolkit for easily adding programmable guardrails to LLM-based conversational systems.”
In this research stack
Features and peers linked from the catalog map — not a second product surface.
Features it supports
- access-vs-authority
- autonomy-levels
- emergency-control
- human-oversight-guardrails
- job-healthcare
- job-security
- r2-13-governance-policy
- r2-16-approval-calibration
- security-identity-governance
- wt-human-policy
Same cell
Stack
Document management
- affinda
- candice-ai
- docplace
- docsum
- docsumo
- docsynecx
- document-one
- docusafe
- docuware
- dokken
- doxsuite
- egnyte
- eigenpal
- eldoc
- eondocs
- extend
- hyperscience
- instabase
- klearstack
- korto
- laserfiche
- m-files
- mindee
- mps-docs
- nectain
- paperoffice
- reducto
- revver
- rossum
- unsiloed-ai
- vertesia
Source
Primary repo github.com/NVIDIA-NeMo/Guardrails · Open Source
technologies/nemo-guardrails/README.md
Research inventory
9 tags · 44 out · 45 in · 3 artifacts · 0 gaps · 33 corpus docs
Catalog tags
- landscape.layer
- Agentic
- landscape.subcategory
- Guardrails
- license_tag
- Open Source
- maps.dm
- present
- maps.features
- 10
- one_liner
- Agentic
- review.depth
- science
- slug
- nemo-guardrails
- title
- NeMo Guardrails
Artifacts
- dm_map · present · technologies/nemo-guardrails/document-management.md
- features_map · present · technologies/nemo-guardrails/features.md
- readme · present · technologies/nemo-guardrails/README.md
Out · alternative_to
- guardrails-aitechnology
- headroomtechnology
- llm-guardtechnology
Out · dm_axis
- affindadm_vendor · dm_vendor
- candice-aidm_vendor · dm_vendor
- docplacedm_vendor · dm_vendor
- docsumdm_vendor · dm_vendor
- docsumodm_vendor · dm_vendor
- docsynecxdm_vendor · dm_vendor
- document-onedm_vendor · dm_vendor
- docusafedm_vendor · dm_vendor
- docuwaredm_vendor · dm_vendor
- dokkendm_vendor · dm_vendor
- doxsuitedm_vendor · dm_vendor
- egnytedm_vendor · dm_vendor
- eigenpaldm_vendor · dm_vendor
- eldocdm_vendor · dm_vendor
- eondocsdm_vendor · dm_vendor
- extenddm_vendor · dm_vendor
- hypersciencedm_vendor · dm_vendor
- instabasedm_vendor · dm_vendor
- klearstackdm_vendor · dm_vendor
- kortodm_vendor · dm_vendor
- laserfichedm_vendor · dm_vendor
- m-filesdm_vendor · dm_vendor
- mindeedm_vendor · dm_vendor
- mps-docsdm_vendor · dm_vendor
- nectaindm_vendor · dm_vendor
- paperofficedm_vendor · dm_vendor
- reductodm_vendor · dm_vendor
- revverdm_vendor · dm_vendor
- rossumdm_vendor · dm_vendor
- unsiloed-aidm_vendor · dm_vendor
- vertesiadm_vendor · dm_vendor
Out · maps_to
- access-vs-authorityfeature
- autonomy-levelsfeature
- emergency-controlfeature
- human-oversight-guardrailsfeature
- job-healthcarefeature
- job-securityfeature
- r2-13-governance-policyfeature
- r2-16-approval-calibrationfeature
- security-identity-governancefeature
- wt-human-policyfeature
In · alternative_to
- guardrails-aitechnology
- headroomtechnology
- llm-guardtechnology
In · dm_axis
- affindadm_vendor · technology
- candice-aidm_vendor · technology
- docplacedm_vendor · technology
- docsumdm_vendor · technology
- docsumodm_vendor · technology
- docsynecxdm_vendor · technology
- document-onedm_vendor · technology
- docusafedm_vendor · technology
- docuwaredm_vendor · technology
- dokkendm_vendor · technology
- doxsuitedm_vendor · technology
- egnytedm_vendor · technology
- eigenpaldm_vendor · technology
- eldocdm_vendor · technology
- eondocsdm_vendor · technology
- extenddm_vendor · technology
- hypersciencedm_vendor · technology
- instabasedm_vendor · technology
- klearstackdm_vendor · technology
- kortodm_vendor · technology
- laserfichedm_vendor · technology
- m-filesdm_vendor · technology
- mindeedm_vendor · technology
- mps-docsdm_vendor · technology
- nectaindm_vendor · technology
- paperofficedm_vendor · technology
- reductodm_vendor · technology
- revverdm_vendor · technology
- rossumdm_vendor · technology
- unsiloed-aidm_vendor · technology
- vertesiadm_vendor · technology
In · in_stack
- agentic-guardrailsstack · alternative_slot
In · maps_to
- access-vs-authorityfeature
- autonomy-levelsfeature
- emergency-controlfeature
- human-oversight-guardrailsfeature
- job-healthcarefeature
- job-securityfeature
- r2-13-governance-policyfeature
- r2-16-approval-calibrationfeature
- security-identity-governancefeature
- wt-human-policyfeature
Corpus tags
- category
- Agentic → Guardrails
- dedication
- open-source
- feature
- access-vs-authorityautonomy-levelsemergency-controlhuman-oversight-guardrailsjob-healthcarejob-securityr2-13-governance-policyr2-16-approval-calibrationsecurity-identity-governancewt-human-policy
- kind
- map_edgetech_featurestech_quotetech_readmetech_science_sourcetech_section
- needs_deepen
- true
- quality
- duplicate_quotesok
- section
- Document management mapEvidenceFeatures mapGitHub mapHow it worksLimits & failure modesLinksScientific notesWhat it isWhen to use / skipWhy it matters here
- slug
- nemo-guardrails
- source_id
- arxiv-2310-10501nemo-colangnemo-github
- technology
- nemo-guardrails
Corpus documents (33)
map_edge · 10
- nemo-guardrails → access-vs-authority
- nemo-guardrails → autonomy-levels
- nemo-guardrails → emergency-control
- nemo-guardrails → human-oversight-guardrails
- nemo-guardrails → job-healthcare
- nemo-guardrails → job-security
- nemo-guardrails → r2-13-governance-policy
- nemo-guardrails → r2-16-approval-calibration
- … +2 more
tech_features · 1
- NeMo Guardrails · features
tech_quote · 7
- NeMo Guardrails · arxiv-2310-10501
- NeMo Guardrails · arxiv-2310-10501
- NeMo Guardrails · nemo-colang
- NeMo Guardrails · nemo-colang
- NeMo Guardrails · nemo-github
- NeMo Guardrails · nemo-github
- NeMo Guardrails · nemo-github
tech_readme · 1
- NeMo Guardrails
tech_science_source · 3
- NeMo Guardrails · arxiv-2310-10501
- NeMo Guardrails · nemo-colang
- NeMo Guardrails · nemo-github
tech_section · 11
- NeMo Guardrails · Document management map
- NeMo Guardrails · Evidence
- NeMo Guardrails · Features map
- NeMo Guardrails · GitHub map
- NeMo Guardrails · How it works
- NeMo Guardrails · Limits & failure modes
- NeMo Guardrails · Links
- NeMo Guardrails · Scientific notes
- … +3 more