- A tool performs an operation: search, read, compare, validate, execute, write, or call an attached plugin.
- A skill is focused guidance for how to solve a class of problems with the available tools.
- A plugin adds tenant-specific tools, skills, MCP servers, or a package of related capabilities.
What the model can see
RevoEngine uses progressive capability loading instead of inserting every available integration and playbook into every turn.
Attached packages can preload compact information about their skills and tools. A deferred capability remains discoverable without putting every complete schema or instruction body into the initial context.
Skill loading
The built-in skill catalog exposes concise summaries. The Agent can search it and load selected skills in one bounded request. Required skill dependencies can load with the selected workflow. Full skill content is supplied to the model only when loaded. Persisted activity keeps the useful identity and outcome metadata rather than copying private instruction bodies into normal conversation history. When an Instance Administrator enables Eager skill loading, supported read operations can attach one selected, reviewed skill directly to the returned evidence. This reduces a follow-up round trip. It does not load every suggested skill, consume a data cursor, trim authoritative evidence, or bypass capability admission.Plugin loading
- Explicitly attached plugins are available to the current thread or Agent subject to current policy.
- Package and skill plugins can describe their dependent capabilities without duplicating schemas or credentials.
- Tools marked for deferred loading are advertised compactly and expanded only after selection.
- Active auto-discovery plugins are deferred candidates, not an always-loaded global toolset.
- A package dependency that is disabled or unavailable keeps the dependent workflow unavailable.
Research loading
Public web research and RevoEngine Docs are optional read capabilities. They are model-visible only when enabled for the instance and admitted for the current execution. RevoEngine Docs is disabled by default; public web access is likewise policy-controlled. Research output is external evidence. It cannot prove current tenant configuration, resource access, deployment state, or a successful RevoEngine operation. Private application data, source code, identifiers, credentials, and signed URLs must not be sent as public research queries.Product knowledge and current application state
The Interactive Agent combines a compact product model with focused guidance and current evidence. RevoEngine Docs is an opt-in preview that provides read-only search and page retrieval from the official documentation. An Instance Administrator enables it under Settings → AI Agents → Runtime when users need current product guidance or citations. It stays disabled by default because documentation lookup can add several network calls and increase response latency. Documentation explains product behavior and design choices. It does not prove that a capability is enabled in your instance, that you have access, or that an operation succeeded. Exact runtime contracts and authorized resource state must still be inspected before execution. If documentation is unavailable or conflicts with the current instance, the Assistant should identify the limitation rather than invent support. Use generic product terms for public documentation research. Private records, source code, identifiers, and credentials do not belong in those queries. See Choose your approach for the shared product model.Platform-aware capability families
Exact availability depends on the current identity, target, Agent or thread configuration, and instance policy. These are public capability families, not internal operation names.
These families describe product capabilities, not unconditional permissions or a guarantee that every capability is available in every thread.
Breadth without overclaiming
RevoEngine can combine model reasoning with live, authorized application context, execution evidence, generated runtime types, persistent workspace artifacts, controlled web research, and instance-specific plugins. This makes it broad for the lifecycle of a RevoEngine application: design, build, debug, operate, and improve. It is not currently positioned as a universal, unrestricted computer operator. It deliberately does not bypass tenant roles, Secret masking, target ACLs, deployment state, or review controls, and it does not promise a general-purpose shell to every Agent. The right measure of its strength is the quality of a verified outcome in the permitted scope—not the size of a capability catalog.Reference tokens
The Assistant uses stable references to keep a name tied to the correct platform object and, where needed, a specific version. In the UI these appear as readable pills rather than raw identifiers. References support safer follow-ups:Skill libraries
Skills provide reviewed guidance for recurring work without changing the underlying permission model. Organizations can use built-in and approved plugin skills to standardize how work is scoped, validated, documented, and handed off. Skill areas can include:Application architecture
Components, dependencies, low-code review, shared libraries, and schema evolution.
Data
Database configuration, row reads, controlled mutations, and query patterns.
Endpoints and runtime
Endpoint contracts, validation, execution, and managed runtime behavior.
Automation and operations
Jobs, schedules, events, webhooks, execution history, and incident investigation.
Storage and identity
Files, upload flows, secrets, users, groups, roles, and preferences.
Agent capabilities
Plugin creation, tool composition, handoffs, memory, and workspace-aware execution.
Authorization and approval
Before an operation executes, RevoEngine evaluates:- Is the capability enabled for this runtime and instance?
- Is it permitted for the current thread or Agent?
- Can the current identity access the target?
- Is the call read-only, write-capable, destructive, executable, or external?
- Does the capability or plugin policy require approval?
- Are the arguments valid for the current operation contract?
Activity and artifacts
The execution timeline can show relevant operations, approval requests, results, and retained artifacts. RevoEngine applies masking, access control, and retention policy to this evidence. Secrets and private platform implementation are not included in normal shared-thread output.Ask for capability-aware work
You usually do not need to name an implementation-specific operation. State the outcome and evidence requirement:Plugins
Add governed tenant-specific and external capabilities.
Approvals and plans
Control when tools pause before side effects.