Case study: a technical video series for Cognee
At a glance
- Client: Cognee, an open-source AI memory platform that turns documents and conversations into a knowledge graph, giving AI agents long-term memory across sessions.
- Work: a scripted tutorial series for Cognee's YouTube channel, produced in 2026.
- Output: fourteen episodes. Nine are published on Cognee's channel as of September 24, 2026, and five more are awaiting publication.
- Reach so far: more than 8,000 views across the nine published episodes. The most watched, on running Cognee with AWS Bedrock, has passed 2,200.
- Deliverables per episode: a script, a demo repository verified end to end, a 16:9 graphics deck, and the finished edit.
What the series covers
Most episodes pair Cognee with a tool its audience already uses: Claude Code, OpenClaw, Slack and Notion, Ollama for fully local models, and AWS Bedrock. Around those sit a concept episode comparing RAG with AI memory, a Python quick start that builds a company knowledge base in two function calls, and a companion episode that builds the same thing on Cognee Cloud. The episodes share a fictional company, Relay, and a recurring cast, so the series reads as one story rather than a stack of unrelated demos.
The most-watched episode so far:
How each episode was built
The series ran on one production system.
- The demo comes first. Each episode starts as a working demo repository whose README lists every step in exact order, plus two teardown scripts: one that restores the starting state between takes, and one that removes everything when the episode is done. Recording starts only after the README runs end to end from a clean machine, so a viewer who follows it gets the result they saw on screen.
- Every claim is checked against the source code. Where the documentation and the installed package disagreed, the source won, and the issues found along the way went to Cognee's team as bug reports. The series doubled as a developer-experience audit.
- The simplest path wins. Demos use the library's documented, recommended commands and APIs rather than clever hand-rolled equivalents, because the viewer has to be able to repeat them.
- Honest over scripted. When a beat did not work on record day, it was cut or fixed, never faked.
- A package, not just a video. Each episode shipped as a script, a production plan, a graphics deck with one asset per slide and timecoded speaker notes, and the verified demo repo. Every link carries a shared campaign tag, so Cognee can attribute visits from the series.
Results
- More than 8,000 views across the nine published episodes as of September 24, 2026.
- Most watched: Cognee on AWS Bedrock (more than 2,200 views), the Python company-brain quick start (more than 1,500), and the fully local knowledge graph with Ollama (more than 1,100).
- Five more episodes are produced and awaiting publication.
Cognee keeps its signup and activation figures private, so this page reports only public numbers.
What this means for your product
A series works best when most episodes pair your product with a tool your developers already use, because each episode reaches the people searching for that tool. It works only if the demo is verified before the camera rolls, because a tutorial that fails on the viewer's machine costs you the user it was meant to win.
A single video up to 10 minutes is $4,000; a series of six or more is $3,000 per video, and every price includes the script, the verified demo repo, the graphics, the edit, and one round of revisions. Details are on the Courses and Video page.
Updated September 2026.
Tell me what your developers need to learn: thecodetraveler@gmail.com.