Personal AI Knowledge Assistant Project: Build a Useful Memory

A personal AI knowledge assistant project is a deliberate setup: a model, a curated set of sources, a list of what the assistant can and cannot do, and a record of what it has found and decided. Done well, you end up with a repeatable workflow rather than a one-off chat. Done without a persistent record, each session resets and you spend the first few minutes re-explaining the project.

The session record is the part most people skip. Tools like fenbs keep decisions, tasks, and where each session left off outside any one AI app, so the next conversation starts from a known state rather than from scratch. For related project types, AI design assistants for project work covers a similar setup for design teams.

What the project involves

A knowledge assistant project has four parts: the model you use, the sources it draws on, the tools it can reach, and the record it maintains. Each part has failure modes. A model with no source list will invent plausible facts. A source list with no access control will expose private material. A tool list with no limits lets the assistant act on the wrong thing. A project with no record resets every session.

Start small. Pick one subject area and one or two source types. Expand only when the narrow version works well enough to trust. A narrow, reliable assistant is more useful than a broad, unpredictable one.

Deciding what the assistant can see

List your sources before you connect anything. Separate public web pages from private documents, and separate reference material from working notes. The assistant will treat everything it can reach as equally trustworthy, so the quality of its answers depends directly on the quality of what you put in front of it.

• Public documentation and official help pages: reliable for facts about how a product or system works.
• Your own notes and drafts: useful context, but watch for speculative or outdated material.
• Emails and messages: check what is personal or confidential before adding any.
• Third-party roundups and summaries: treat these as pointers to primary sources, not sources themselves.

Write the list down. When you check an answer later, you need to know where the assistant could have found it. A source you forgot you added is a source you cannot audit. The AI for project managers guide covers a similar discipline for project knowledge.

Limiting tool access

Every tool the assistant can call is a decision it can make on your behalf. A search tool lets it look things up. A write tool lets it save or send things. Grant the minimum set of tools the project actually needs, and set the scope as narrowly as the tool allows.

If the assistant only needs to read documents, do not give it write access. If it only needs to search one folder, do not give it access to all your files. Review the tool list whenever the project scope changes. Tools that were safe at the start may become risky once the assistant is working in a new area.

Keep a note of which tools are connected and why. When you move between AI apps, that note tells you what the new one needs and which tools the previous setup had that you do not want to replicate automatically.

Checking sources before trusting outputs

Ask the assistant to name the source for each factual claim before you use the answer. A response that cites a specific page, section, or date is easier to check than one that gives a confident summary with no trail. If the cited source says something different from the answer, the answer is wrong regardless of how well it reads.

Build source checks into your workflow rather than doing them only when something looks suspicious. An assistant that always cites its sources trains you to look at them. An assistant that never does trains you to trust it instead.

Record sources that have worked well and sources that produced wrong answers. That record helps you refine the list and gives a future session a head start. For research-focused projects, the AI research assistant project guide covers source management in more depth.

Keeping state between sessions

Most AI apps do not preserve useful project state between sessions by default. A new chat knows nothing about the decisions made in the last one. You end up re-explaining the project, correcting the same misconceptions, and rediscovering the same sources.

The fix is an external record the assistant reads at the start and updates at the end. That record holds the project goal, the decisions that govern it, the tasks in progress, and where the last session left off. It survives a browser refresh, a model change, and a move to a different AI app.

How to do this with fenbs

1. Create a free account. Add the knowledge assistant as a project on your personal kanban board.
2. Write the project goal and add the tasks the assistant will work through. Keep the first version short: one goal, three to five tasks, and where things stand.
3. Record the rules that govern the project in Decisions and rules: which sources are trusted, which tools are allowed, how to handle a conflicting answer. These rules appear first in the AI context every session.
4. Connect your AI app over MCP from Integrations. Ask the assistant to read the project context before each session begins.
5. At the end of each session, ask the assistant to save progress: what it completed, what it found, and what the next session should start with. Keep that record current and the project stays useful indefinitely.

Handing the project to a different assistant

If you switch AI apps mid-project, the work you recorded stays in fenbs. Connect the new app over MCP and ask it to read the same project context. It finds the goal, the rules, the tasks, and where the last session left off. You do not have to explain the project again.

Check that the new assistant recognizes the constraints before it acts. Ask it to summarize the rules and the current next step. If it misses a constraint, add it to Decisions and rules before continuing. A constraint in the record is a constraint every future session will see. The how to use AI agents in project management guide covers how to apply this across larger projects.