User guide / Agentic chat
Install Playground GitHub ↗ Book Cloud

Self-host with the docs, star Enverif on GitHub, or book a Codefreex discovery call for managed cloud and custom plugins · MCP · skills.

Star on GitHub Book discovery call

Agentic chat workspace

Chat is Enverif's primary operating surface. A chat thread stores conversation context and user-facing defaults, while every submitted user turn creates a separate durable AgentRun with its own execution snapshot, tool steps, approvals, cost accounting and terminal result.

Persistent thread defaults

Each conversation can remember four defaults:

  • primary agent;
  • model connection/provider account;
  • model ID;
  • execution effort: Fast, Standard or Deep.

The composer exposes Keep for this chat and Just this message behavior. Keeping a selection updates the thread defaults. A one-message override is stored with the run but does not silently rewrite the thread.

Fast/Standard/Deep is provider-neutral. Enverif maps effort to supported provider reasoning controls where they exist and uses bounded execution guidance where they do not. It never sends a provider parameter that the selected adapter does not support.

In-place browser transport

Sending a message is an AJAX/Fetch operation from the chat composer. The browser does not navigate to a transport endpoint. On the first turn Enverif creates the thread, updates the address bar to the canonical thread URL with history.replaceState, replaces the transcript with server-rendered safe HTML and polls the durable run status in place. The composer is locked while the current run is non-terminal so duplicate turns cannot race the same thread.

If JavaScript is unavailable, the normal HTML form fallback still redirects to the canonical thread URL after creating the durable run. /chats/send is not a public route in 1.3.

In Shared/Compatibility mode, interactive sends also schedule a bounded post-response queue kick so a database-queued run can begin without waiting for the next cron minute. Cron/Web Cron remains required for unattended work.

Searchable history

The sidebar groups active chats into Today, Yesterday, Previous 7 days and Older. Search matches conversation titles/content. Threads can be renamed, archived or deleted. Archived threads remain recoverable from the history surface until explicitly deleted.

Deleting a thread removes its Enverif-private attachment files as well as the thread records. Durable audit/run records are governed by their own retention semantics and are not rewritten merely because a title changes.

Structured @ context

Use the + context picker or type @ in the composer to select structured context:

  • @agent
  • @plugin
  • @skill
  • @workflow
  • @lead
  • @campaign

These selections are IDs stored in run context, not decorative prompt text. Workspace authorization is applied before the IDs are accepted. A tagged plugin grants only its declared actions for that run; a tagged skill adds that skill to prompt context; a workflow/lead/campaign tag supplies scoped business context.

Attachments

Chat files are stored on the private Laravel disk, not directly under public/. Access requires the same authenticated user and workspace. The composer accepts common images and business documents subject to configured MIME/size limits.

  • text-like files are supplied as bounded text context;
  • supported images are supplied as provider image content;
  • files that the current provider cannot ingest remain visible as secure attachment metadata rather than being exposed publicly.

Do not use chat attachment storage as a general document vault. Large knowledge bases should be implemented through a dedicated skill/plugin or indexed source.

Run status and final responses

While a run is active the chat status can show states such as:

  • Thinking
  • Using Apollo / another tool
  • Waiting for approval
  • Final

These are operational states, not hidden chain-of-thought. Enverif does not expose private model reasoning. The final assistant record stores execution metadata such as agent, provider/model, effort, run ID and selected context so the answer can be traced back to the durable run.

The Stop action cancels the run tree, including delegated child runs, using the same persisted cancellation path as the run console.

Approvals

External writes such as email send/reply pause in Approvals by default. Approval resumes the exact persisted tool step; denial returns a denied tool result to the agent. Autonomous external writes require an explicit agent/workflow policy choice and still remain audited.