Projects, Files, and Memory

A project is the shared folder for a client, brand, or story. It keeps that work's files, memory, and media lists together, so the Medialyst agent can start with the context your team already has.

Every organization starts with a General project, shown in the app as Default project. When you begin a campaign, choose the project before sending the first chat message. The project is fixed after the chat starts, and the media list created from that chat stays with it.

Create or open a project

If Projects is not shown in your app navigation, it is not available for your organization yet.

1. Select Projects in the app navigation.
2. Open a project card, or select Create project if your plan has room for another project.
3. Enter a name and an optional description, then select Create project.

You can also select the project pill below the campaign chat composer and choose New project. Make this choice before sending the first message; the picker locks when the chat starts.

From a project page, select Start a new chat to begin with that project already selected. The page also collects every media list created in the project. See Find Journalists for the ways you can add journalists after the first list is built.

Add project files

Project-file upload is available on paid plans.

1. Open Projects, then open the project.
2. In Files, drop files into the upload area or select Choose a file.
3. Wait for each file to show Ready before relying on it in chat.

Each project file or chat attachment whose upload starts counts as one attempt, even if the upload or extraction later fails.

Spreadsheets are not project files. To add journalists from a CSV, use Add journalists → From CSV instead; the Find Journalists guide covers that workflow.

Read the extraction receipt

After Medialyst reads a file, its card shows a short generated description and an extraction receipt titled What Medialyst kept from this file. The receipt can include:

• The number of PDF pages and readable characters
• Whether a long document was truncated
• Detected section headings and a text preview
• Warnings about anything that could not be kept

The description helps the agent decide which file is relevant. The receipt lets you check the text Medialyst can actually use. If a file shows Couldn't read or Reading didn't finish, remove it, export a fresh text-based copy, and upload it again. Password-protected documents and PDFs over 100 pages are not supported.

What the agent reads

Project memory is loaded automatically for every chat turn in the selected project. The agent also receives an index of ready project files with their titles, descriptions, amount of readable text, and truncation status. It opens or questions the relevant files when it needs their contents; it does not paste every full file into every turn.

Relevant project files are available throughout the chat, but the agent opens their contents as needed rather than before every response. If a particular brief, white paper, coverage record, or prior list should shape the work, name it in your prompt. Project context also guides the ranking of journalists when Medialyst builds the next list.

You can ask evidence-backed questions in campaign chat. For example:

In the uploaded launch brief, what proof points support the enterprise angle? Quote the supporting language.

For this kind of question, the agent can search ready project files and check its answer against verbatim evidence from the extracted text. If the files do not state the answer, it should say so instead of guessing.

Write and edit project memory

Viewing and editing the memory file is a paid-plan feature.

1. Open Projects, then open the project.
2. In Memory, select Edit.
3. Add or revise the Markdown, then select Save.

The default memory file has four useful sections:

## About the client
- Durable facts about the company, market, geography, and spokespeople.

## Ranking preferences
- Trade outlets rank above national news for product launches: their readers are the client's buyers.

## Never
- Never pitch non-US media: the client serves US buyers only.

## Pitch voice
- Lead with the evidence, not a product claim.

Memory can contain up to 20,000 characters. Keep one fact or rule per bullet and include the reason behind ranking preferences and hard exclusions. Your edited wording wins: the agent can append new learnings, but it does not rewrite a line you edited.

Teach a durable preference in chat

In a campaign chat for the project, state the durable rule plainly and include its reason. For example:

Remember for this project: US business and technology trades rank above national general news on product launches because they reach the client's buyers.

The agent saves one durable line as part of the turn. It does not open a separate approval step or ask you to confirm the wording first. After a successful save, look for the small Noted in Project name memory receipt below the chat response. Paid users can follow its view link to inspect the memory file.

The agent keeps campaign-only instructions in the current campaign instead of memory. Facts about one journalist or one story belong in the list or a coverage record, not in a durable client rule.

Teach from a scored journalist row

This feedback flow is available on paid plans and on rows with a completed Journalist Profile result.

1. Open a media list and select a journalist's Journalist Profile result.
2. Under the relevance score, select Poor fit.
3. In What should Medialyst learn?, explain the durable reason in one sentence.
4. Select Save to memory.

The panel shows Noted in Project name memory when the note is saved. Good fit is an acknowledgement only and does not add a memory line.

You can make up to 60 scored-row memory save attempts per person, per organization, per UTC day. The allowance resets at midnight UTC.

Use the reason to express the rule you want reused, not just your verdict on this row. For example, write “Assigning editors rank above editors-in-chief for trade stories: they are more likely to commission them,” rather than “wrong score.” Medialyst routes the learning into the appropriate memory section and keeps source information for the correction.

This action does not change the journalist's current score. The learning applies when you build a later list in the same project.

Sharing, provenance, and plan limits

Projects belong to the organization, not to one user. Organization members work from the same project files, memory, and media lists. Chat-added memory and scored-row corrections carry source and date information; on paid plans, the memory reading view marks those additions as made by the agent or by a user. File receipts record what was extracted from each upload.

Project allowances include the Default project:

If support has raised only your organization's project-count limit, that changes how many projects you can have. It does not unlock file uploads or memory access on the Free plan.

Best practices

• Use one stable folder per body of context. A long-running client usually deserves one project; a confidential or unusually distinct story may deserve its own.
• Put source material in files. Upload client briefs, prior lists exported to a supported format, known-coverage records, press releases, white papers, and proof assets.
• Put judgment in memory. Record geography, outlet tiers, send-order preferences, and pitch voice. Include the reason so the agent can apply the rule to edge cases.
• Separate preference from exclusion. “Industry trades rank first because their readers are buyers” guides ranking. Use Never only for a genuine no-exception rule. Uploading a file or naming an outlet does not automatically exclude anyone.
• Pair records with a rule. A coverage file can show who wrote about the client; memory should explain whether repeat writers rank higher for continuity or lower because the client wants new bylines.
• Check the receipt. A file marked Ready with a sensible description and extraction preview is more useful than a scan with no readable text.

Once the next list exists, use Research Journalists to inspect the evidence behind each result and Tables to organize, filter, and share the list.