AI that can't lie,
leak, or go rogue
O-Lang is the open orchestration language for governed AI workflows. Declare your resolvers, enforce your intent, audit everything — cryptographically, at the kernel level.
// O-Lang — Bank Balance Workflow Workflow "Bank Balance Check" with user_question, customer_id Allow resolvers: - @o-lang/bank-account-lookup - @o-lang/llm-groq Step 1: Ask @o-lang/bank-account-lookup "{customer_id}" Save as account_info Step 2: Ask @o-lang/llm-groq "{user_question}. Balance is {account_info.balance}." Save as response Return response
Workflow domains in v1.1
Cryptographic audit coverage
Hallucinations in governed scope
Open source license
Governance baked in,
not bolted on
Every O-Lang workflow passes through the kernel — a runtime that validates, governs, and signs every execution before returning output.
Write your workflow
Define steps, resolvers, and governance rules in a .ol file. Declare what's allowed. Declare what's prohibited. That's your contract.
Kernel enforces governance
The O-Lang kernel validates every resolver call against your allowlist, checks intent scope, runs safety filters, and blocks prohibited actions — before execution.
Cryptographic audit trail
Every execution is SHA3-256 fingerprinted and ED25519-signed by the kernel. Your outputs are verifiable, tamper-evident, and ready for compliance.
Everything a governed
AI runtime needs
Resolver Allowlists
Every workflow declares exactly which resolvers it may call. Any unlisted resolver is rejected by the kernel at runtime — no exceptions.
@governance.allowed_resolversVerified Intent
Declare scope and prohibited actions. The kernel enforces both before and after execution — zero tolerance for scope creep.
__verified_intentCryptographic Audit
Every execution produces a SHA3-256 fingerprint, chained to the previous, and signed with ED25519. Full deterministic proof.
sha3-256 · ed25519Symbol Resolution
Steps compose with typed symbols. Each resolver declares its output schema — downstream steps receive validated, typed data, never raw LLM output.
$step.symbol.fieldSafety Policy Engine
Built-in safety filters block hallucinations, PII leakage, speculative outputs, and prompt injection — configurable per domain.
safety_policyResolver Registry
Publish, version, and discover resolvers at registry.o-lang.dev. Every resolver is signed and schema-validated before listing.
@o-lang/resolver-nameWrite and run
.ol files in your
browser
No install. No config. Connect to the O-Lang cloud kernel and execute governed workflows live.
- Monaco editor with O-Lang syntax highlighting
- Real-time execution trace — see every resolver step
- Live safety logs and cryptographic audit output
- Save and share workflows with a permanent URL
// ICU Admission — Healthcare workflow workflow icu_admission_v1 { @governance { allowed_resolvers: [ "@o-lang/icu-admission", "@o-lang/llm-groq" ] scope: "team-a-icu-only" prohibited_actions: [ "unauthorized_procedure", "ward_transfer_without_auth" ] } step admit = @o-lang/icu-admission({ patient_id: $input.patient_id, ward: $input.ward, team: $input.team }) }
Discover and publish
governed resolvers
Every resolver in the registry is schema-validated, version-pinned, and signed. Drop it into any workflow with a single import.
@o-lang/bank-account-lookup
finance
Secure account balance and transaction lookup with PII validation and multi-jurisdiction audit compliance.
@o-lang/icu-admission
health
ICU bed management, team-scoped admission, and HIPAA-compliant patient data handling.
@o-lang/llm-groq
llm
Fast, governed LLM inference via Groq. Output schema-validated and hallucination-filtered.
@o-lang/notify-telegram
utils
Send governed notifications to Telegram channels or users. Rate-limited, audit-logged, and allowlist-enforced.
@o-lang/semantic-doc-search
utils
Vector-based semantic document search with scoped retrieval. Prevents out-of-context document access.
@o-lang/js-olang-tester
utils
Unit and integration test harness for O-Lang workflows. Simulates resolver chains and validates governance policy enforcement.
@o-lang/python-olang-tester
utils
Python test harness for O-Lang workflows. Available on PyPI. Validates resolver behaviour and policy compliance from Python environments.
@o-lang/legal-extract
legal
Extract-only legal document analysis. Blocks legal advice output under GDPR, UK Companies Act, US Federal, and other jurisdictions.
@o-lang/llm-openai
llm
OpenAI GPT-4o resolver with output schema enforcement and rate-limit-aware execution.
@o-lang/pii-validator
finance
Global PII validation across 40+ identity formats: SSN, NIN, BVN, NHI, TFN, Aadhaar, and more.
Governed AI for
high-stakes domains
O-Lang was designed for industries where AI errors are not just inconvenient — they are dangerous, illegal, or catastrophic.
Finance
Balance inquiries, transaction classification, fraud detection — all with PII validation, hallucination prevention, and CBN-ready audit trails.
Healthcare
ICU admissions, clinical triage, ward management — with team-scoped access, HIPAA filters, and zero-tolerance prohibited procedure enforcement.
Legal
Contract extraction, NDA review, clause identification — extract-only mode blocks legal advice output, protecting against unauthorized practice of law.
Everything you need
to build governed AI
From installing the kernel to publishing your first resolver — step-by-step guides, API reference, and real-world examples.
Your AI should be
as accountable as you are
Join researchers, engineers, and enterprises building the next generation of governed AI systems on O-Lang.