Knowledge and Agents

Home · Projects and workspace · Structured outputs, evidence and freshness

Goal

This page explains how to upload or import documents, review the metadata visible in Knowledge, and then use the project context in Agents.

Knowledge: what you can do

Observed capabilities:

• search within project contents;
• manual upload;
• data refresh;
• import from approved sources;
• view import history;
• sort, filter and paginate results.

Supported file types

The explicitly accepted types in the interface include:

• PDF;
• DOCX, ODT, RTF, TXT, Markdown;
• CSV, XLSX, XLSM, ODS;
• JSON, XML;
• PPTX, ODP;
• HTML / HTM;
• EML, EPUB;
• PNG, JPG, JPEG, BMP, TIFF, HEIF.

Import from approved sources

The import mechanism depends on ingestion providers validated at the platform level, then made available to the project.

What the import selector shows

Depending on the source, the interface may display:

• the source name;
• a readable source label;
• the date of the last import;
• a freshness or health status;
• an explanation of blockage when the source is unusable.

When an import is blocked

An import option may be unavailable due to:

• entitlement;
• permission;
• policy;
• availability status to check;
• missing or not ready project binding.

Import history

The import history allows you to verify at minimum:

• the displayed provider name;
• the run status;
• freshness after import;
• the Trace ID;
• volumes discovered, imported, ignored or marked as incomplete;
• the end date when available.

Reading the document list and search

Visible document states

The document table exposes at least the following states:

• Indexed;
• Ingesting;
• Failed.

A visible document that is not yet searchable is often still in the Ingesting phase.

Metadata visible in search results

Search results may display:

• title;
• snippet;
• score;
• section;
• freshness badge;
• source system badge;
• source label badge;
• Open source link;
• search Trace ID.

Practical reading:

• source label usually identifies the visible provider or ingestion origin for that evidence, such as manual, SharePoint, ADF, Blob, Confluence, or Jira;
• source system helps you recognize the human-readable system or workstream behind the evidence, e.g. Knowledge, Schedule, Finance, or Operations;
• the search Trace ID is mainly a diagnostic, support, and audit handle rather than a day-to-day business key.

Use source label and source system together when you need to judge provenance quickly before reusing a snippet.

Agents: page role

The Agents page leverages the project context in a guided conversation.

Confirmed functions:

• list of available agents;
• summary card of the selected agent with kind, scope, status and covered domains;
• conversation area;
• structured outputs and evidence items;
• sessions saved in the browser;
• creation and deletion of custom agents according to rights;
• voice input in Agents if the environment still supports it.

ProPM Agent proactivity

Proactivity in ProPM Agent does not mean an agent acts alone outside the project. The observed behavior is more controlled: the platform detects situations to monitor, surfaces them as signals, can prepare a digest, propose a draft or a governed action, then leaves the user or approver to decide the next step.

Observed triggers

Proactivity can be fueled by:

• insufficient freshness;
• a contradiction between evidence;
• an operational blockage;
• a planning pressure or costs visible in the portfolio;
• unfinished executions;
• a follow‑up requiring dissemination, validation or relaunch.

Where proactivity becomes visible

• Workspace: signal feed and items to review;
• Governance, decisions and actions: digests, notification drafts, action requests;
• Portfolio command center: project comparison by indicators;
• AI Journal: traceability of the decision chain.

What proactivity does not do alone

• it does not bypass roles;
• it does not bypass policies;
• it does not automatically publish outside the application without a governed path;
• it remains explainable by evidence, freshness and visible statuses.

How to pick an agent quickly

When multiple cards seem plausible, start with the covered domain, then verify scope and status.

Field	What to look at first	Useful reflex
covered domains	topics announced on the card	match the request to the agent whose domain actually covers the work you need
scope	agent visibility scope	confirm whether the agent is project-scoped or visible more broadly
status	current availability of the agent	prefer agents that are active for the selected project before escalating
kind	whether the card is standard or custom	treat kind as descriptive context, not as the first selection criterion

In practice, search the roster by name or domain first. Prefer an active agent for the selected project, and keep transverse or custom agents for cases where their domain coverage really fits the request.

How to work with agents

1. check the active project;
2. choose the appropriate agent;
3. formulate a usable request with objective, scope and expected format;
4. review the structured output before any reuse.

The interface also allows searching the agent roster to reach the right specialist faster.

Recommended beginner path

For a first use without ambiguity:

1. ask a simple question in Agents;
2. read the summary of the structured output first;
3. then verify evidence, freshness and confidence if the answer will be reused;
4. open Reports & artifacts only if you want to turn that result into a governed, reviewable or publishable document.

What All projects really means in Agents

When the custom-agent creation dialog exposes All projects, it means:

• the agent is still created from the current project;
• it can remain visible across all projects accessible by the same account;
• it follows your account from one project to another;
• it does not create a merged multi-project data space;
• it does not replace the active project at run time.

Even an All projects agent still runs in the active project at the moment you send the chat message.

From local browser chat to PM Documents, DOCX, and XLSX

Agents do not jump directly to a final distributed file. The observed flow is more progressive:

1. the current chat first remains browser-local;
2. sending the message launches a run in the active project;
3. the run produces a structured output;
4. depending on the flow and your rights, this output can expose lineage links, an artifact, or a PM Doc;
5. review then continues in Reports & artifacts / PM Documents;
6. the reviewed document can finally be Add to knowledge, Download, or Publish according to governance.

What structured output means

A structured output is the structured result produced by the agent run. It can include:

• a summary;
• sections;
• evidence or citations;
• freshness and confidence information;
• traceability identifiers.

It is not yet, by itself, a final published document.

What lineage means

Lineage links together:

• the run;
• the structured output ID;
• the context snapshot ID;
• the artifact and its version;
• the PM Doc when a governed document exists.

This lineage helps you understand where a deliverable came from and which context was used.

Sessions and local memory

Sessions and local memory

Chat history is local to the browser. In practice:

• continuity saved is tied to the project + agent pair in the current browser;
• a session is stored locally on the machine used;
• changing browser, profile, machine, or clearing local storage can cause a saved thread to disappear;
• the first prompt usually serves as the session title, which aids resume;
• do not treat the browser as a shared central archive.

Local browser vs shared platform

Keep this distinction simple:

• chat continuity and the remembered project are browser-local conveniences;
• documents, imports, artifacts, approvals, and publications belong to the shared platform.

Custom agents

Creating custom agents depends on the level of rights. Key points to remember:

• not all users can create an agent;
• the creation dialog recalls the active project context before saving;
• visible fields cover at minimum Name, Role / instructions and Scope;
• choose Project only by default, unless the environment explicitly exposes a wider scope;
• if the All projects option is available, the agent is created from the current project but can remain visible in other projects for the same account;
• if a multi‑project agent does not appear elsewhere, first check the same account and access to the target project;
• deletion of a custom agent remains controlled;
• a custom agent remains governed by the project and platform rules.

Quick choice example

Need	Recommended choice
nominal usage, large team, predictable behavior	standard agent
specific instructions, team tone, specialized workflow	custom agent
usage limited to a single project context	Project only
same agent reusable in multiple projects of the same account	All projects if the option is truly exposed

Voice input in Agents

This documentation no longer describes a separate voice entry point in the Workspace. When voice input is still available, it is limited to the Agents experience and depends explicitly on browser support for speech recognition APIs. Its absence does not block the nominal path, which remains text input.

If voice input is not available:

• use text input;
• check browser microphone permissions;
• distinguish a browser limitation from a local environment constraint;
• always review the text added in the composer before Send;
• if the mic button is unavailable or an incompatibility message appears, switch immediately to text input.

Practical voice input cues

Symptom	First check	Useful reflex
mic button missing or inactive	support for SpeechRecognition / webkitSpeechRecognition	switch immediately to text input
mic permission denied	browser or OS mic permission	re‑authorize mic then restart listening
incomplete transcription	text added in the composer	review, correct and complete before Send
occasional error during listening	browser or current session state	stop capture, resume with keyboard, then try again later

Text input remains the default fallback path. It allows work to continue without interrupting the run due to a local browser constraint.

Best practices

• load reliable and recent sources into Knowledge before a major analysis;
• review import history when a managed source seems missing;
• explicitly request evidence or citations if the decision must be justified;
• if a response contains stale, conflicting or unavailable states, open the page Structured outputs, evidence and freshness before publishing.

Next

• Structured outputs, evidence and freshness
• Governance, decisions and actions
• Reports, AI Journal and traceability
• Maintenance, support and FAQ