All Docs
FeaturesSaaS FactoryUpdated March 11, 2026

Agent Activity Transparency & Live Explainability Feed

Agent Activity Transparency & Live Explainability Feed

Available since: v1.0.125

The pipeline detail view now exposes a structured, real-time feed of everything an agent is doing — its tool calls, reasoning steps, and the exact code diffs it is writing. This gives operators and enterprise buyers full auditability of autonomous deployments without leaving the dashboard.

Overview

SaaS Factory agents operate autonomously, but every action they take is now visible in the Pipeline Detail page. The feed is powered by the existing agentLogs table and Server-Sent Events (SSE) streaming infrastructure, rendered through an updated live-pipeline-detail.tsx component.

There are three distinct layers of information surfaced in the feed:

LayerWhat you see
Reasoning highlightsThe agent's chain-of-thought — why a feature was selected, why an approach was chosen
Tool-call collapsiblesEvery tool invocation (file reads, searches, writes, API calls) with inputs and outputs
Inline diff renderingSyntax-highlighted code diffs for every file the agent creates or modifies

Reasoning Highlights

Each agent decision that involves non-trivial reasoning is surfaced as a highlighted block in the feed. This includes:

  • Feature selection rationale — why the AI architect decomposed the work the way it did.
  • Approach decisions — why one implementation strategy was chosen over another.
  • Risk or compliance flags — if the agent paused due to a detected issue, its reasoning is shown.

Reasoning blocks are visually distinct from action output so they are easy to scan during a review.

Tool-Call Collapsibles

Every tool the agent invokes appears as a collapsible row in the feed. By default the row shows:

  • Tool name (e.g. read_file, write_file, search_code)
  • A one-line summary of the call (e.g. the file path or search query)
  • Elapsed time

Expanding a row reveals the full input parameters and the tool's output or return value. This makes it straightforward to verify that the agent read the right files and produced the expected results before writing any code.

Inline Diff Rendering

When the agent writes or modifies a file, the diff is rendered directly in the activity feed with:

  • Syntax highlighting matched to the file's language.
  • Unified or side-by-side diff layout.
  • File path and change summary (lines added / removed) in the row header.

Diffs are streamed incrementally as the agent writes, so operators can watch code being produced in real time.

Architecture Notes

  • No new back-end services. The feature is built entirely on the existing agentLogs table and SSE streaming endpoint.
  • All structured rendering happens in live-pipeline-detail.tsx on the client.
  • Log entries are tagged by type (reasoning, tool_call, diff) so the renderer can apply the correct layout without post-processing.

Use Cases

Enterprise Approval Workflows

Enterprise buyers who need to approve autonomous deployments can open the pipeline detail page, review the agent's reasoning and every file diff, and approve or reject without needing to inspect the underlying repository.

Debugging Failed Runs

When a pipeline fails, operators can scroll through the tool-call history to identify exactly which step produced an unexpected result — without needing to parse raw log files.

Auditing Completed Pipelines

The feed is persisted via agentLogs and remains accessible after a pipeline completes, providing a full audit trail for compliance purposes.

Related Pages