From markdown to connected knowledge

Context rots

Every new session, an agent starts from zero. The human remembers a little — but fidelity decays fast, and chat transcripts become unreadable after a week. The decisions, the alternatives considered, the constraints that shaped the work — they scatter. Temper keeps them on a throughline that survives the session it was made in.

Markdown with frontmatter

Every piece of knowledge in temper is a markdown file with YAML frontmatter. The frontmatter carries structure — type, status, relationships, mode, effort. The content carries the thinking. Together they form a file that's simultaneously human-readable, git-trackable, and natively understood by language models.

The vault is a directory on your machine. Browse it in any editor, in Obsidian, or on GitHub. No database, no proprietary format. Import existing docs with temper add — temper extracts markdown and adds the frontmatter structure.

Every session, a loop

Sessions aren't arbitrary — they follow a recurring shape. You warm up with recall from the vault, do the work, reach the decisions that moved it forward, and record what happened before closing the tab. The vault grows on every loop. The colors of each step match the doc types they produce: blue for the warm-up's research recall, gold for the decision, green for the session record. Only the work itself is uncoloured — it's the doing between bookends.

Every idea knows its neighbors

Files in your vault aren't isolated — they're connected. A task relates to a goal. A session produces decisions. Research informs the next task. Temper tracks these relationships and builds a knowledge graph that grows with every session. When you search, you don't just find a file — you find the constellation of ideas around it.

The dashed lines are the interesting ones — decisions feeding back to inform future tasks, research shaping goals. The graph captures the connections that context rot erases. When an agent runs temper warmup, it reads not just the current task but the neighborhood of decisions and sessions that shaped it.

Your vault, everywhere

Temper Cloud is Postgres-native — a single source of truth for your knowledge base, with pgvector powering semantic search across everything you've written. Your local vault stays as markdown files on your machine. The cloud makes them searchable, syncable, and accessible from any device.

Changes flow, conflicts resolve

temper sync uses a manifest-based protocol. Your vault keeps a local manifest that records the last-known state of every resource — a three-way comparison between the file on your machine, the manifest's record, and the server's content. Non-conflicting changes merge automatically at the paragraph level using Rust-native diffing. Genuine conflicts get annotated in a .conflict.md file for human resolution.

Your vault can also live in a git repo, sync to Obsidian, or coexist with anything else that reads files. Temper's sync is self-contained — it doesn't depend on or interfere with external version control.

Everything resolves to markdown

This is the invariant. Whatever happens in the cloud — search indexing, vector embeddings, knowledge graph construction, team access, agent queries — it all resolves back to markdown files with frontmatter in your vault. The files are the source of truth. The cloud is a lens on them. If temper disappeared tomorrow, your knowledge would still be there: readable, portable, yours.