Skip to main content
RevoEngine keeps three kinds of state separate. This makes Agent behavior easier to understand, govern, and resume: Personal preferences and instance-wide policy are configuration, not memory. See Agent configuration and personalization.

​
Conversation and run continuity

An Interactive Agent thread keeps the messages, plans, approvals, actions, and artifacts needed for the current collaboration. An Autonomous Agent run additionally keeps durable lineage and checkpoints so authorized work can continue after a pause, retry, or scheduled wake-up. Use this state for live progress. Do not copy every intermediate action into memory.

​
How runtime memory works

Interactive Assistant memory remains a bounded digest for its authorized USER context. Historical WORKSPACE records do not add an automatic runtime memory scope. These digests are soft context: they never override the current request, access controls, approval policy, or verified action results. Autonomous Agent AGENT memory is one text value in the Memory tab, limited to 20,000 tokens. Its current text is included in the Agent instructions for every model request. The memory tools read and update the same value; older typed Agent records are retained for historical evidence and are not automatically recalled. Keep memory compact. A larger value increases model input on every request. Store large or exact material as workspace files and retain only the durable convention or reference in memory.

​
Memory writeback

An Agent can update its memory when it discovers stable knowledge that will matter later. Updates are version-aware and use one of three operations:
  • append — add a new durable fact or convention;
  • edit — change a specific existing part;
  • replace — replace the complete digest after loading its full current value.
If the digest changed after it was read, the update must be refreshed instead of silently overwriting newer knowledge. Good memory is concise and remains useful after the current task ends:
Do not save credentials, access tokens, raw reasoning, full documents, transient task progress, or large action outputs. Store exact material in Storage and keep only the durable convention or reference in memory.

​
Search and deeper recall

Interactive Assistant memory retains its governed digest and deeper recall behavior. Autonomous Agent memory tools operate on the current configured text. They do not search historical typed records as an implicit second memory source. Use conversation history and workspace files when exact prior work is needed.

​
Agent workspaces

Every successfully created Autonomous Agent has a workspace root:
  • Managed — RevoEngine creates and governs a dedicated workspace.
  • Existing — an operator links an authorized Explorer folder as the Agent root.
The root is the default location for Agent-generated files and an enforced boundary for workspace operations. Linking an existing folder is an authorization decision: review its contents, service-account access, and ACLs before making it the Agent workspace. Interactive Agents create ordinary Storage files under the current user’s authorization. Autonomous Agents use their assigned service account and workspace policy.

​
Workspace writes and approvals

Writes inside an enforced Agent workspace can proceed without an extra per-action confirmation when the configured policy allows them. Writes outside that boundary remain ordinary Explorer operations and keep their normal permission and approval rules. The workspace boundary does not make content public. Upload, download, preview, archive, and folder operations still require authentication and authorization.

​
Outputs and completion

Reference a generated file only after the platform has persisted it. A durable output can include its Storage entry, name, media type, kind, size, and source activity when available.
Do not store secret values in memory, workspace guidance, generated reports, or Agent messages. Keep secrets in the RevoEngine Secrets service and expose them only through approved runtime bindings.

Agent Harness

See how state, capabilities, policy, and evidence are assembled for execution.

Agent configuration

Separate platform policy, personal defaults, and Agent configuration.

Autonomous Agents

Configure the identity and runs that use these stores.

Storage

Learn how managed files and folders behave outside Agent-specific workflows.
Last modified on October 5, 2026