> ## Documentation Index > Fetch the complete documentation index at: https://docs.canary.effectiveai.app/llms.txt > Use this file to discover all available pages before exploring further. # Key concepts > What each part of the Effective app is for and when to use it: chats, outcomes, Mission Control, ContextBase, Vault, agents, skills, apps, procedures, records, assets and data sources. This page describes key concepts that are helpful for using Effective. ## Chats and Outcomes ### Chats **Chats** are where you work with the agent: ask, research, draft, analyze. You can share a chat with other users, groups or your team. Chats started within a Mission Control are not accessible outside it, and you can restrict one further to just you or specific people using the sharing icon in the chat box (a lock when the chat is private). * **Use it when:** you want an answer or a draft now and will steer it as you go. ### Outcomes **Outcomes** are a kind of chat with a goal: the agent keeps working until it's done, checking in when it needs your input. Each outcome is driven by a Captain: an agent that breaks the work down and delegates it, much like a program manager. The Captain keeps going after you close the chat: it reviews the outcome once a day, wakes up when something changes (a reply, a finished task, a new message), and can set up repeating checks, such as a daily look for new filings. That's why outcomes work well for work that needs tracking over time. Its page shows the goal, the latest update, questions waiting on you, outputs (documents, records, links) and a history of activity. * **Use it when:** the work has several steps, takes days, or needs ongoing tracking, and seeing progress is important. > [!TIP] > > Use a chat when you're driving; use an outcome when you want the agent to drive to a result. If > unsure, use a chat. ## Mission Control **Mission Controls** are used for organizing work and setting access boundaries. By default, resources created in a Mission Control are not accessible outside it. Most of them can be given broader access by their creator; chats are a notable exception and always stay inside their Mission Control. Mission Control admins control who has access. The creator is an admin by default. Mission Controls also decide which skills, apps and procedures are in play. Install one from its catalog and it's available to everyone working there, within their access. New chats pick up installed skills automatically. Each Mission Control draws its background knowledge from one or more ContextBases. Start a fresh one, or connect one you already have, such as your team's shared ContextBase, so several Mission Controls can share the same knowledge. * **Use it when:** you're working on a sensitive project and want to restrict access to a specific set of people. * **Use it when:** a team needs the same skills, apps and procedures on hand without setting them up in every chat. ### ContextBase **ContextBase** stores the agent's context for a Mission Control: decisions, guidelines, definitions and how-tos. It should be mostly text and markdown files. Think of it as a special kind of vault whose files are already in the agent's hands when a chat starts, so it can use them right away. It works best when actively curated: keep it current and trim what's stale. When the agent saves something to ContextBase, it shows you the change first and only publishes once you approve. The system can optionally require specific people to approve changes. * **Use it when:** it's written knowledge the agent should always have on hand. For notes and guidance, prefer it over the Mission Control Vault. * **Use the Mission Control Vault instead when:** the files are large or aren't text, such as big datasets, spreadsheets or scanned PDFs. A large ContextBase makes every chat in the Mission Control slower to start. * **Try:** "Save our decision on the new territory definitions to ContextBase." or "What does our ContextBase say about appetite for coastal properties?" ## Vault **Vault** is where files live. Files are also called artifacts. Choose a vault by who should see the file. * **Personal:** only you can see it. Use it for drafts and scratch files. * **Team:** everyone in your organization can see it. Use it for org-wide files. It's read-only inside a Mission Control; to add or change files, start a global chat outside any Mission Control. * **Library:** reference documents Effective provides, such as public rate and form filings from SERFF. You can read them but not write to them. * **Mission Control Vault:** files for one Mission Control, visible to its members. Also see [ContextBase](#contextbase). **Try:** "Save this spreadsheet to the Mission Control vault under reports/." or "Find the latest homeowners filing for this carrier in the Library." ## Agents and Skills ### Agents **Agents** are AI workers with a job. A general agent runs every chat. You can also make focused agents with fixed instructions for one job, and anyone in the Mission Control can pick them from the agent dropdown in the chat box. There are two ways to make one: * **In ContextBase:** ask the agent to add one to the Mission Control's ContextBase. It's a short instructions file, so it's easy to review and change. By default it reads and answers, without running code or editing files. * **As an app:** ask in chat for an agent app. Choose this when the agent needs its own tools, saved data or connections to other systems (see [Apps](#apps)). * **Settings:** either kind can be allowed to run code. ContextBase agents can also offer starter prompts, and app agents can be set to use a specific model. * **Use it when:** you want a dedicated chat that sticks to one job, such as a guided intake for people who shouldn't need to know which skill to use. * **Try:** "Add an agent to our ContextBase that reviews broker submissions against our appetite guide." ### Skills **Skills** are reusable know-how any agent can pick up: the steps, standards and even scripts for one kind of task. They're the most common way to teach the agent something, and they can run code. * **Use it when:** you've explained the same steps more than twice, or you keep pointing the agent at the same instructions file in the vault. Start here if you're unsure. * **Try:** "/\ on the attached file" or "Turn what we just did into a skill for this Mission Control." > [!TIP] > > A skill adds to the everyday agent, which keeps everything else it can do, including running code. > An agent replaces it with a narrower one that sticks to one job. So start with a skill, and make > an agent only when you want to narrow things down. ## Apps **Apps** are tools with their own screens, data and integrations. You build one by asking in chat. Each app comes with: * **Its own database and file storage**, so data is kept between chats. * **Secrets** for connecting to outside services. * **Access to Effective data**, such as filings and records. * **API keys**, so outside systems can call the app. Apps don't run on a schedule by themselves; use a [procedure](#procedures) for that. Every app published to a Mission Control also appears in the agent dropdown. Picking it starts a chat with the app's own agent, which follows the app's instructions and uses only the app's tools. You don't need to pick it to use those tools: the everyday agent in that Mission Control can already use them. * **Use it when:** people need a screen to click through, data has to be kept outside of chats, or the work connects to another system. * **Try:** "Build me an app where underwriters log submissions and see a status board." ## Procedures **Procedures** are fixed workflows that run the same steps every time, with a history of every run. Each step can be a bit of code, an agent, or a person: a person step pauses the run until someone reviews it or fills something in. Procedures can be tuned to save cost, since not every step needs an agent and each agent step can use a preconfigured model. A procedure's page has: * **Procedure, Outline and Graph:** the steps, as a document or a diagram. * **Runs:** every run with its inputs, results and status. Start a run here, or have the agent run it with you. * **Triggers:** start runs automatically on a schedule, when a file lands in a Vault folder, or when an event happens. * **Secrets:** for connecting to outside services. To make one, describe the process in chat, or do it once with the agent, and ask it to turn that into a procedure. The agent writes and tests the steps and publishes when you say so. * **Use it when:** a process must be repeatable and auditable, such as monthly reporting or intake checks. If the agent can use judgment on the steps, a skill is lighter. * **Try:** "Turn what we just did into a procedure." or "Run the monthly rate-monitor procedure for Q3." or "Make this checklist a procedure that runs every Monday." ## Records **Records** are the structured things your organization keeps track of, such as products, filings and accounts. Each kind of record (a record type) has its own fields, and people and agents read and edit the same records. On the Records page you can: * **Browse** each type as a table with search, filters and saved views, kept for yourself or shared with everyone. * **Open a record** to see its overview, documents, related records, and an activity history of every change and comment. * **Request approvals** from people or groups, narrow who can see a record, and see the outcomes working on it. * **Add record types:** install built-in types from the catalog, or create a custom type, for example by starting from a spreadsheet. The agent can do the same from chat. Ask it to look up, create or update records, set up a new record type, or import rows from a spreadsheet. * **Use it when:** you track many of the same kind of thing and want fields, filters and history, not a pile of documents. * **Try:** "Create a record for each carrier in this spreadsheet." or "Set up a record type for surety bonds with the obligee, bond amount and expiry date." or "@\[record name] what changed since the last filing?" ## Assets **Assets** are files with extra labels: the Mission Control they're published to, what kind of deliverable they are (report, tracker, presentation, or a link to an outside resource), and any tags. Some are kept up to date by a procedure. Publishing turns a working file into something the team can find, filter and @-mention. * **Use it when:** a file is done and others should rely on it. Drafts and working files stay in the Vault. * **Try:** "Publish this comparison as a tracker asset tagged Q4." or "@\[asset name] update the Texas numbers." ## Data sources **Data sources** are outside datasets the agent can look things up in, such as: * SERFF filings * EDGAR * Legal records * Regulatory documents * NAIC statements There are more, and new data sources added to the Effective platform show up too. Your organization picks which ones are on by default, and you can change them for a chat from the + button in the chat box. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.