Core Concepts
How sources, feeds, digests, and objectives work together.
How to Build Projects in Cortex
Cortex projects consist of three basic steps:
- Connect sources. Sources provide the underlying content for your project.
- Create feeds to route your source content through. Each feed acts like a filter.
- Define digests as your final outputs. Digests turn feed content into something you can use.
Objectives
Objectives are how Cortex ensures that AI agents are doing what you want them to do. You should think of objectives as your project's North Star. When building projects in Cortex, there are three types of objectives: Project Objectives, Feed Objectives, and Digest Objectives.
Separate objectives make the project easier for both you and AI agents to understand. Project, feed, and digest objectives tell different parts of the workflow what they are trying to accomplish, so it is easier to see what is working, what needs to change, and where to make improvements.
Gives the project a single goal that the rest of the workflow supports.
Example: Monitor the most important AI research and product updates.
Tells the feed what to keep and what to skip.
Example: Keep major model launches, papers, benchmarks, and safety findings.
Describes the output you want.
Example: Produce a daily research monitor briefing with the key updates and why they matter.
Once you understand the basic workflow from sources to feeds to digests, and how to define objectives for each layer, you already have enough to start building simple projects in Cortex. If you want to see what that looks like in practice, explore the Examples page.
Deep Dive: Projects, Sources, Feeds, and Digests
Projects
A project is the top-level organizational container within Cortex that groups related feeds and sources under a shared objective. A project gives a Cortex workflow one coherent job. The sources, feeds, and digests inside it should all support the same operating goal rather than acting like unrelated automations.
A project objective is the top-level statement of what a Cortex project is meant to accomplish across its feeds and digests. A strong project objective defines the domain, the broader use case, and the kind of signal the system should ultimately help surface.
A good objective for a project states what all of the sources, feeds, and digests in the project are generally meant to help achieve.
Example: Help Acme's marketing team stay up to date on important developments across the real estate industry, social media, and competitor landscape.
Sources
A source is an external content provider that Cortex connects to for periodically retrieving raw content. Sources are the intake layer. They answer one question: where does the raw material come from? An RSS feed, X account, listing page, or JSON endpoint can all be valid sources if they are the right way to access the signal you need.
A source should describe access and coverage, not editorial judgment. If you find yourself writing detailed relevance rules into a source choice, that logic probably belongs in the feed objective instead.
Feeds
A feed is a configured ingestion channel governed by an objective that determines which incoming content items are relevant and how they are summarized. Feeds are the interpretation layer. They evaluate incoming content items against the feed objective, keep what matters, reject what does not, and summarize the material that passes.
This is where you express inclusion and exclusion criteria. When a feed keeps too much, misses important items, or produces muddy summaries, the feed objective is usually the first place to tighten the system.
A feed objective is the scoped statement of what incoming content should be treated as relevant within a feed. A strong feed objective makes the signal boundary legible by stating what to prioritize, what to exclude or deprioritize, and what kinds of developments are materially worth keeping.
A good objective for a feed states what kind of information belongs in the feed so anything else can be filtered out.
Example: Include information about competitor pricing changes, packaging updates, and customer complaints in the cybersecurity industry so the product marketing team can track what may affect positioning. Filter out general cybersecurity news that is not directly relevant to competitors or customer buying decisions.
Digests
A digest is a configured, recurring output of Cortex that consumes content from one or more feeds and produces a structured deliverable aligned to its objective. Digests are the delivery layer. They take the content your feeds have kept and turn it into a recurring output for a person, coding agent, or other system.
A digest objective should describe the consumer and the job to be done. It should not try to replace feed scoping; it should define how the kept signal should be organized and used.
A digest objective is the statement of what recurring artifact a digest should produce from feed data for its intended consumer. A strong digest objective defines the audience, purpose, and organizing lens of the output so Cortex can shape the result instead of merely restating the feed's scope.
Use Briefing when the output should read like a narrative report, and use Dataset when the output should become a structured table. The Digest Types guides cover the output-specific settings and troubleshooting details.
Feedback and optimization
You can thumbs-up or thumbs-down individual feed items and digest outputs. When you do, Cortex uses that feedback to improve future runs.
Feeds and digests also show their optimization history so you can see what changed and why. Over time, this turns Cortex from a content pipeline into a system you actively tune.
Where to go next
- Source Types to choose the right connector for each intake pattern.
- Examples to see complete Cortex workflows built from these layers.
- Glossary for the canonical definitions behind the main domain terms.
- API Reference for endpoint-level schemas and contracts.