Remember what you decided
Every session builds on the last. Your vault holds the throughline — what's been done, what's up next, what the agent needs to know.
The blank session
You're three sessions into a feature. On Tuesday you chose JWT rotation over session tokens. Wednesday you explored two caching approaches and picked the one with simpler invalidation. Now it's Friday, and the agent has no idea any of that happened.
So you re-explain. You paste fragments of old conversations. You write ad-hoc notes that may or may not be found again. The industry has started calling this context rot — the progressive degradation of an agent's understanding as work spans sessions. The more complex the project, the heavier the tax.
Context that compounds
Temper is a local-first knowledge base that embeds throughline directly into how you work. Not a task tracker competing with Linear. Not a doc tool competing with Notion. A structured vault of markdown files where every goal, task, session, decision, and research thread has a home — and where the connections between them are always visible.
The vault grows richer with each session. Decisions persist. Research accumulates. The agent that starts your next session reads what you decided, what you deferred, and what comes next — not because you re-explained it, but because the vault holds the thread.
Warm up. Work. Save.
Every session follows a rhythm. Start with context. Do the work. Record what happened. The next session reads what you wrote. This is the core loop — and it's what turns a collection of documents into a living history of what you're building.
Find decisions by meaning
You don't remember the filename. You barely remember the date. But you know you made a decision about authentication strategy somewhere in the last two weeks. Temper embeds your vault using local ML models and searches by meaning, not keywords.
Results span decisions, sessions, research, and tasks — all ranked by relevance to what you actually meant, not what you happened to type.
Your workflow, your way
Temper doesn't impose a process. It gives you two building blocks — goals and tasks — and lets you define what they mean for how you work.
The outcome you're building toward. A goal holds the vision and purpose of a feature, a product, a body of work. Tasks and sessions roll up to goals — the throughline runs through them.
A unit of work toward a goal. Every task has a mode — build or plan — and expected effort — small, medium, or large. Code is the common case, but a research proposal, a slide deck, even a novel — the thing you're building is up to you.
When you run temper init, the setup asks how you work. Do you use superpowers? GSD? Something of your own? Temper adapts to your workflow rather than replacing it — it carries the throughline regardless of what tools and ceremonies you prefer.
Everything resolves to markdown
Your vault is a directory of markdown files with YAML frontmatter. Human-readable in any editor. Version-controllable in git. Natively understood by every language model. No proprietary formats, no required tooling to view your own knowledge.
The frontmatter carries the throughline — type, status, relationships. The content carries the thinking. Git carries the history.
Five minutes to your first session
Temper also works as a direct context layer for agents — through the CLI, MCP server, and generated skill files. If your agent can read files, it can use temper.