Tana's AI features are INSANE

Tana Labs adds two AI-native building blocks—AI Fields and AI Commands—that let ChatGPT-style workflows run directly inside a workspace, using the notes, tags, and fields already stored there. The practical payoff is faster knowledge work: definitions can be generated for concepts, missing metadata can be extracted automatically, claims can be stress-tested with counterarguments, research plans can be drafted and refined, and voice notes can be transcribed and rewritten into structured summaries.

Setup starts with enabling Tana Labs via the gear icon, then selecting the AI features option. Using the system requires an OpenAI API key from platform.openai.com, created under “View API keys.” The transcript emphasizes two constraints: API usage costs money, and every AI action sends workspace data to OpenAI—so sensitive information shouldn’t be used unless it’s acceptable to share externally. Once the key is pasted into Tana, AI-enhanced fields and commands become available.

AI Fields focus on enriching existing structured content. For example, a “Concept definition” field tied to a concept node (like “expertise” or “second brain”) can be converted into an AI-enhanced field. A custom prompt is configured to ask for a definition in under 100 words, using Tana’s field-reference syntax with dollar-sign curly brackets to pull in the node name and other field values (such as language or style). A built-in prompt workbench lets users test the prompt against a specific node before saving. After configuration, clicking the AI-enhanced field runs the prompt and inserts the generated definition back into the workspace. The same mechanism can support translation, etymology, and example sentence generation.

AI Commands turn text-based instructions into reusable actions. Commands can be created by converting text into a command node (via command K → “convert to command note”). A key example is “get topics,” where an “ask AI” sub-command extracts topics from a quote and writes the result into a target field (the “topics” field). The workflow can be triggered either by selecting the command from the command line or by adding a button to nodes through Codex configuration so the action runs on demand.

More advanced prompting appears in “countering a claim.” Here, a command evaluates a claim using a role-based prompt (“world famous expert” on specific topics) and produces counterarguments. The transcript demonstrates passing in both the claim text and the topics it relates to, yielding multiple refutations; running the command on opposing claims produces a larger set of pro-and-con arguments.

Finally, the system supports planning and voice-note processing. A “suggest research plan” command adds a new field, generates a step-by-step plan, and outputs search terms for further investigation. It uses a two-step prompting pattern: first draft a plan, then critique and improve it by feeding the initial output back into a second prompt. For voice notes, Tana can transcribe audio (command K → “transcribe audio”) and then run a “summarize voice note” command that produces a title, a first-person past-tense summary, main points, action items, follow-up questions, and potential arguments against—alongside the original transcript context. The overall message: Tana Labs turns a second brain from a storage system into an active assistant for thinking, organizing, and decision support.