Files, Intake & Trail
TensorPM treats files and incoming signals as project context inputs, not as a separate document archive.
The important flow is:
- Files, emails, and raw updates enter Intake.
- Supported items can be summarized or evaluated.
- The Distiller turns relevant signals into reviewable project updates.
- Applied changes appear in Trail.
- Decisions can be recorded as append-only commitments.
File Explorer
The File Explorer is the project-facing file view.
Common actions:
- upload files
- create folders
- open or reveal files in the system file explorer
- rename and delete files
- move items by drag and drop
- attach files to Action Items
- generate AI summaries for supported file types
- include or ignore files for distillation
The File Explorer supports grid and list views. In list view, use column headers for sorting.
Project files stay close to the context where people and agents actually use them.
Project file root
A project can have a file explorer root path. When configured, the File Explorer uses that location as the browsing base.
Use a dedicated project folder when:
- project documents already live in a known folder
- you want TensorPM to reference real working files
- you want local agents to operate near the same filesystem context
Avoid using a broad home directory as the project root. It creates too much noise and increases the chance that unrelated files become visible in project workflows.
Supported AI file behavior
For supported file types, TensorPM can:
- generate file summaries
- use file content in AI chat when attached
- evaluate whether the file contains project-relevant updates
- queue the file for distillation
Unsupported files can still exist in the File Explorer, but they will not produce the same AI workflow.
Attach files to Action Items
Attach files when a task depends on a concrete artifact.
Good examples:
- requirements document attached to implementation work
- invoice attached to a budget item
- design mockup attached to review work
- specification attached to a QA task
This keeps execution connected to source material.
Intake
Intake is the raw signal layer. It aggregates project-relevant inputs that may need evaluation.
Current intake sources include:
- project files
- processed documents
- emails from configured connectors
- raw project updates
The header can show Intake activity. Open it when you want to see what is waiting for review before it becomes structured project context.
Preview and hydration behavior
For large projects, Intake may first show a preview and then hydrate more items in the background. This prevents the UI from blocking while TensorPM loads many pending files or signals.
If counts appear to change shortly after opening Intake, wait a moment and refresh. The full snapshot may still be hydrating.
Distiller
The Distiller is the review agent that processes pending signals.
Its normal loop is:
- Analyze one pending signal.
- Present proposed changes.
- Let the user review the proposal.
- Apply approved project updates through tools.
- Verify the result.
- Mark the signal complete and move to the next one.
This makes project updates controlled rather than silently writing every detected change.
Include or ignore for distillation
Use Ignore for distillation for files that should stay in the folder but should not influence project updates.
Use Include for distillation when a file becomes relevant again.
Typical files to ignore:
- duplicate exports
- old screenshots
- reference material not tied to project state
- temporary files
- raw assets that do not contain decisions or requirements
Trail Updates
Trail -> Updates shows processed changes and history.
Trail keeps context changes traceable when files, AI, or agents update the project.
Use it to answer:
- What changed?
- Which field changed?
- Which source caused the change?
- Was the change produced by distillation or another AI/tool action?
- Are there processed signals that need follow-up?
Trail is especially useful when a project changes through files, emails, AI chat, or agent actions instead of manual editing only.
Trail Decisions
Trail -> Decisions stores first-class commitments.
Decision records include:
- decision text
- decision source
- source actor
- decided timestamp
- status
- rationale
- links to Action Items, risks, or milestones
- supersession chain when a decision changes
Do not rewrite old decisions to hide changes. Supersede or withdraw them so the audit history remains useful.
When to record a decision
Record a decision when it changes project behavior.
Good candidates:
- scope inclusion or exclusion
- timeline commitment
- budget constraint
- architecture direction
- vendor choice
- risk acceptance
- stakeholder agreement
- priority rule
Chat messages are not enough for durable commitments. Put decisions in Trail so people and agents can find them later.
File and Trail operating routine
- Upload or connect source files.
- Attach files to specific Action Items when they are execution evidence.
- Generate summaries for large reference documents.
- Ignore noisy files.
- Process pending Intake through the Distiller.
- Review Trail updates after applying changes.
- Record decisions separately from general context updates.
Troubleshooting
If a file does not appear:
- confirm the project file root is correct
- refresh the File Explorer
- check whether the file is outside the selected root
- verify permissions on the local folder
If distillation does not pick up a file:
- confirm the file type is supported for AI processing
- check whether it was ignored
- generate a summary first for large files
- open Intake and verify the signal is pending
Next steps
- Understand project structure: Project Structure
- Attach work artifacts: Action Items
- Configure connectors and MCP clients: Agent Integrations
- Fix file or AI issues: Troubleshooting