Knowledge Store & Wiki: Team Memory for Humans and AI

Documentation lives in one tool, project knowledge in another — and the AI has access to neither. That changes now.

With the new wiki system and knowledge store, Spedy gets a central memory: for your team, for your agents, for every project.

Wiki: Documentation Inside Your Project

Every workspace can now have its own wiki. Create a Wiki Space, organize it with folders, and fill it with pages — all with a simple editor, right inside Spedy.

What the Wiki Can Do

• Spaces & folders — Hierarchical structure with nested folders. Each space has its own slug and an optional homepage
• Version history — Every change is saved as a version. See who changed what and when, and restore any previous version instantly
• Attachments — Upload files directly to pages. Images, PDFs, whatever you need — stored securely in your storage
• Full-text search — Search across all wikis you have access to. Results show page name, snippet, and folder path
• Project linking — Link a wiki to a project and it shows up as a tab in the project navigation

Two Access Models

You decide per wiki how permissions work:

Inherit from project — The wiki inherits members from the linked project. Project admins become wiki admins, members become editors, viewers stay viewers. Zero configuration.

Manage separately — The wiki has its own member list. Add users or teams with one of three roles: Admin, Editor, or Reader.

Knowledge Store: What the AI Learns from Your Work

The wiki is for humans. The knowledge store is for the AI — and for you.

The knowledge store collects learnings from your day-to-day work and makes them searchable. Every entry has a category, a confidence score, and optional links to tickets, projects, or wiki pages.

Six Knowledge Types

• Solution — How do you solve problem X?
• Preference — Use Nuxt 5, not Nuxt 4
• Pattern — Best practice for recurring tasks
• Error Fix — Specific bug resolution with context
• Convention — Project-wide rules and standards
• Insight — General learnings from work

Hybrid Search

The knowledge store combines two search methods:

1. Full-text search — PostgreSQL-based for exact matches
2. Semantic search — Vector embeddings for conceptually similar results

Results are ranked by relevance and freshness. Knowledge marked as Evergreen — meaning it stays valid over time — is never downranked.

Agents Access It Automatically

This is where it gets interesting. Runner teams and project agents have direct access to the knowledge store via MCP tools:

knowledge.recall — At the start of every task, the agent retrieves relevant project knowledge: everything linked to the current project or ticket, plus all evergreen entries across the organization.

knowledge.search — Targeted search for solutions, patterns, or conventions. Filter by category, tags, or linked resources.

knowledge.store — The agent saves new findings: a bug fix, a discovered pattern, a correction. Complete with confidence score and duplicate detection.

What This Means in Practice

When an agent solves a problem, it stores the solution. The next time a similar issue comes up, it finds it again — without any human intervention. Your team's knowledge grows with every resolved task.

And when you correct the agent? The correction is saved as a Preference with confidence 1.0. The agent won't make the same mistake twice.

How to Set It Up

Wiki

1. Go to Knowledge → Wiki Spaces
2. Click Create Space — choose a name, slug, and access model
3. Optional: Link the space to a project
4. Create folders and pages using the tree navigation on the left

Knowledge Store

1. Go to Knowledge → AI Knowledge Base
2. Create entries manually or let agents learn automatically
3. Mark important entries as Evergreen
4. Link entries to projects, tickets, or wiki pages for context

Why Both Belong Together

The wiki is your structured knowledge: processes, specs, onboarding guides. The knowledge store is your operational knowledge: solutions, patterns, corrections.

Together they form your team's memory — and your AI agents' memory. The more you document and the more your agents work, the better the system gets.

We're curious how you'll use it. Give it a try and let us know what your knowledge store learns.