Interactive Agents
Work interactively in persistent threads with streaming progress, references, attachments, plans, and approvals. The current UI and API label this surface Assistant.
Agentic coding
Investigate implementation, understand dependencies, produce code, validate it, and apply governed changes.
Autonomous Agents
Run durable, service-account-backed workers from direct requests, inbox items, schedules, and events.
Plugins
Extend the runtime with component tools, MCP servers, reusable skills, and capability packages.
Agent Harness
See how RevoEngine grounds, governs, continues, and verifies Agent work.
Compare coding agents
Understand where RevoEngine is stronger and where repository-first tools are a better fit.
One runtime, two ways to work
Both surfaces use the same Agent Harness and platform-aware capability model. Read operations can inspect authorized platform objects and operational evidence. Mutations remain subject to identity, role, capability policy, safety classification, and approval policy.
Assistant remains the current user-interface and API name for interactive threads. This documentation uses Interactive Agent as the product category and keeps Assistant where an exact screen, setting, or API contract requires that label.
An agentic system for the application lifecycle
A repository-first coding agent primarily begins with source files and a terminal. RevoEngine begins with the governed application model that builds and runs backend applications and services. It can connect engineering intent to platform objects, execution evidence, and authorized operations across the backend lifecycle.
This does not make an Agent an unrestricted administrator. It operates only within the identity, capabilities, policies, and targets admitted for the current task. See RevoEngine and coding agents for an explicit comparison with repository-first tools.
Governed work lifecycle
The execution timeline shown in the UI distinguishes progress, capability activity, approval requests, plan steps, and final messages. Large details and generated files can be retained as governed artifacts.Grounded by platform state
The Harness works with stable references to the objects you can access—for example a Component, Endpoint, database, Agent, Job, or stored file. It can inspect current details, read historical snapshots, compare versions, and analyze dependencies before proposing or applying a change. This grounding is especially important for development work. The Assistant should resolve the intended target, inspect its current contract, and validate the outcome instead of generating code against an assumed schema.Governance by default
RevoEngine applies several independent controls:- Identity and roles determine which tenant data and operations are available.
- Capability policy limits what a thread or Agent may use.
- Approval policy decides whether a side effect can run immediately or must pause for review.
- Planning policy can require a reviewable plan before implementation begins.
- Plugin policy governs each installed capability, including external MCP tools.
- Audit evidence preserves relevant operations and outcomes without exposing Secrets or private platform implementation.
Choose your next guide
Configure Agent behavior
Separate tenant-wide policy, personal Interactive Agent defaults, Autonomous Agent configuration, memory, and workspaces.
Use an Interactive Agent
Learn the thread, reference, attachment, sharing, and execution-detail workflow.
Review plans and actions
Separate execution mode, planning policy, and approval policy.
Configure an Agent
Define durable responsibilities, triggers, limits, and operational ownership.
Understand memory and workspaces
Decide where conversation context, learned knowledge, and generated files belong.