AgentGuard
SecurityMake agent tool calls subject to explicit policies, limits and inspectable audit records.
Reference service Explore ↗
AI infrastructure / security / governance
Intelligence without blind spots.
Seven engineering systems for teams that need to see what their agents do, prove why they did it, and hold the boundary while autonomy grows.
07Portfolio systems
06Next development branches
DOIPublication records
OpenSource you can inspect
We Do Care GlobalEngineering in motion / 01
Accountable intelligence.
01 / The stack
One inspection layer.
One inspection layer.
Six control surfaces.
Each system owns a single control question. OmniGuard sits in front of all of them, screening every request before it reaches an agent, a tool or a model.
Inspection layer
Request screening, policy checks, session risk and governance headers.
00
Open ↗
OmniGuard
Unified gateway in front of every other system. Development preview, private source.
AgentGuardAgentVaultHybrid-RAGAgent EvalFoundationCentarPharma OS
02 / The systems
Six systems.
Six systems.
One clear purpose.
Start from the problem you need to solve. Every row links to the source, the publication record and the current development branch.
FoundationCentar
GovernanceExplore governance-first foundations for operational AI deployment and accountability.
Next branch Explore ↗Enterprise Hybrid-RAG
RetrievalConnect enterprise knowledge with hybrid lexical and dense retrieval, fused through RRF.
Next branch Explore ↗Agent Eval
EvaluationMake agentic workflows easier to inspect through evidence-driven metrics and execution assessment.
Next branch Explore ↗AgentVault
AssuranceBring tool registration, secret management and bounded execution into a clearer control layer.
Next branch Explore ↗Pharma Intelligence OS
Regulatory intelligenceExplore a dedicated workspace for pharmacovigilance and regulatory intelligence workflows.
Review required Explore ↗
03 / The approach
Build the capability.
Build the capability.
Keep the control.
The useful question isn't only what an agent can do. It is how you know it did the right thing, under which policy, with which evidence.
-
Set the boundaries.
Make tool access, rate limits, credential scope and human review explicit before expanding autonomy. A policy that is not written down is not a policy.
-
Connect the evidence.
Ground responses in knowledge, record every decision with a hash-linked trail, and keep the inspection layer in front of the systems you actually run.
-
Earn the next step.
Move from demonstration to deployment through tests, operational checks and stated acceptance criteria. A passing test is evidence, not a certification.
Your starting point / Security
AgentGuard
Start with the boundaries around what your agents can do. Review policies and audit behaviour before connecting privileged tools.
Explore AgentGuard ↗
Enterprise offer / The gateway
OmniGuard.
OmniGuard.
Inspect before acting.
One gateway in front of the whole portfolio. Every request is screened, checked against policy, scored for session risk, and returned with governance headers the calling system can act on.
- Layered injection and trust-boundary inspection before any model call
- Tool-policy evaluation with explicit ALLOW / BLOCK decisions
- HMAC-chained central audit with a verifiable chain
- Six integration adapters for the systems in this portfolio
Request inspectionIllustrative
injection_checkpass — no instruction override pattern
trust_boundarypass — source marked untrusted
vault_lookuppass — 2 scoped credentials
policy_enginereview — tool not in allowlist
session_riskpass — score 0.11
governanceawaiting human approval
DecisionBLOCK · PENDING APPROVAL
04 / The evidence
Trust is inspectable.
Follow the work from source to publication. Code, permanent publication records and a research identity that ties them together.
The engineering
Read the source.
Inspect the organization: project code, documentation, test suites and development activity.
GitHub organization ↗ The publicationTrace the work.
Publication record, published source and current development branch for every system in the portfolio.
Publication index ↗ The peopleKnow the creator.
Connect the software portfolio with Emir Perla's public research identity and publication history.
ORCID profile ↗
33%
of annual corporate revenue pledged to humanitarian causes
We Do Care Global connects open engineering with a humanitarian purpose. The pledge is written down so it can be checked, not just stated.
Founder-level commitment; not a report of donations made. Read the humanitarian constitution ↗
05 / Start a conversation
What are you
What are you
building toward?
Bring a concrete challenge. We'll find the right starting point for your AI infrastructure — and say plainly if the answer is that you don't need all seven systems.
emir@wedocare-global.com ↗Sarajevo, Bosnia and Herzegovina · Open to global collaboration
07 / About
We Do Care Global
We are a global initiative governed by Emir Perla, AI Workflow & Automation Engineer, dedicated to transparent, accountable, and governed AI infrastructure. Our mission is to provide open engineering systems that teams can inspect, verify, and build upon.