01
// IDENTITY
Real actors
User, agent, tenant, project, environment, task, approval chain. No anonymous automation.
◆
Open-source agent control plane
PRIVATE SYSTEMS · POLICY FIRST
Run open-source agents.
Keep corporate control.
Weles Enterprise is the governed execution layer for companies that want open-source agents inside real systems: identity, permissions, tool brokering, sandboxing, approvals, and replayable audit.
[ local agents ]
[ brokered tools ]
[ full audit ]
REC
WLS-ENT-0001 / AGENT GOVERNANCE FEED
[ 01 ]
Protocol
The agent is not the security boundary.
The model may plan, reason, and request action. It never decides what it is allowed to do. Authority lives in Weles: policy, scoped secrets, tool execution, approvals, and audit.
02
// POLICY
External authority
Permissions are evaluated outside the model by role, data class, tool risk, and resource.
03
// BROKER
Mediated tools
Agents see a reduced tool surface. Every call is checked, scoped, logged, and returned as evidence.
04
// AUDIT
Replayable work
Prompts, retrieved docs, denied calls, approvals, tool I/O, artifacts, cost, and errors are reconstructable.
[ 02 ]
Why
Open-source agents are inevitable. Ungoverned agents are not.
// companies want local control
// teams want open tools
// security wants hard boundaries
Enterprises will not standardize on one hosted black-box agent. Teams will bring their own open-source agents, frameworks, models, and tools. Weles gives those agents a controlled operating layer before they touch company systems.
This is not a chatbot with SSO. It is infrastructure for autonomous work: a tool firewall, a policy engine, an approval system, a sandbox broker, and an audit trail built for internal systems.
Use the agent you trust. Run it where your data is allowed to live. Let Weles decide what it may see and do.
[ 03 ]
Runtime
Every task runs like a controlled operation.
One task. One scoped runtime. One explicit policy envelope. Credentials are temporary, network access is bounded, and the final artifact has a trail.
$ weles run agent://sre --project payments-api
> LOAD IDENTITY: kacper / sre-oncall
> FILTER TOOLS: metrics.read, logs.read, k8s.staging.exec
> BLOCK: k8s.production.mutate
> REQUEST APPROVAL: production diagnostic command
> STORE TRACE: WLS-AUDIT-8F31
STATUS: CONTAINED ✓
[ 04 ]
Control Plane
Admin UX for agent permissions, not chatbot settings.
Security and platform teams need a matrix: who can run which agent, what it can read, what tools it can call, which actions need approval, and what model boundary applies.
Tool Permissions Matrix
Configure each open-source agent by role, project, environment, risk class, and resource.
Tool
Eng
SRE
Sec
Ops
GitHub: open pull request
ALLOW
ALLOW
ALLOW
DENY
Kubernetes: production mutate
DENY
APPROVE
APPROVE
DENY
CRM: update account notes
DENY
DENY
DENY
ALLOW
Restricted data export
DENY
DENY
APPROVE
DENY
Prompt
User delegates incident triage for payments-api.
Retrieval
Runbooks filtered before model context.
Tool Call
Metrics allowed; production command escalated.
Artifact
Diagnosis and approval trail stored for replay.
[ 05 ]
Agent Profiles
Bring the agents. Weles supplies the operating discipline.
Engineering, SRE, security, data, and operations teams can run different open-source agents under the same enterprise control plane.
Code Agent
Reads repos, explains code, creates patches, opens PRs, and observes CI without production authority.
SRE Agent
Reads logs, metrics, and runbooks, then proposes remediation behind approval gates.
Security Agent
Enriches alerts, triages vulnerabilities, creates tickets, and preserves suppression controls.
Data Agent
Queries approved datasets, respects row-level security, and blocks restricted exports by default.
Ops Agent
Reads CRM context, drafts external communication, updates notes, and requires confirmation to send.
[ 06 ]
Deployment
Packaged for teams that cannot leak the boundary.
Weles is strongest where agent work meets regulated data, internal systems, private infrastructure, and audit obligations.
Team Server
Shared gateway, SSO, basic policy, central logs, and brokered connectors.
Self-Hosted
Kubernetes, customer VPC, customer database, customer secrets manager.
Private Models
Azure OpenAI, Bedrock, local Qwen/Llama, or classification-based routing.
Regulated Mode
CMK, private endpoints, strict retention, SIEM export, and replay reports.
Air-Gapped
No external telemetry, explicit model boundary, egress control, offline operation.
Open-source agent power. Enterprise containment.
Weles gives companies a hard operating layer between agent autonomy and internal systems: policy, brokered tools, scoped credentials, approvals, isolation, and audit.