How Martium works, start to finish
Martium is a background process, not a chat window you have to remember to open. This page walks through exactly what it captures, how it connects that activity into something queryable, and what it looks like when you actually ask it something.
Capture. Connect. Recall.
Three stages, running continuously in the background, in this order, every time.
Capture
Martium quietly indexes your activity as you work — no manual logging, no "remember to write this down." It reads:
- GitHub commits, diffs, and pull request discussion
- Jira / Linear tickets and their status changes
- Confluence and Notion docs you actually touch
- Local file changes — diffs, not full snapshots
- Debugging sessions: terminal output, stack traces, the actual back-and-forth of fixing something
- Meeting audio, captured from system audio during calls you're already in
Wherever possible, Martium reads accessibility text directly from the app you're using instead of taking a screenshot — it's faster, cheaper, and doesn't need to look at your screen to understand what's on it. Screenshots are a last-resort fallback for genuinely opaque content, like a diagram or a video frame, and raw images are discarded once the useful text is extracted.
Not everything gets deep analysis. A lightweight pass looks at metadata first — app name, window title, whether this is a new task — and only escalates to a closer read when it's actually worth the cost. That's what keeps capture fast and unobtrusive instead of a constant drain on your machine.Connect
Raw activity isn't useful on its own — a commit and the ticket it closes and the debugging session that led to it are three disconnected facts unless something ties them together. Martium links your personal activity into Deimos, your private timeline, and merges it at query time with Phobos, your team's shared knowledge pulled from connected tools.
Nothing is force-fit into a rigid schema. If you jump between three tasks in an afternoon, Martium keeps them as separate threads instead of blending them into a confusing single record — task isolation is deliberate, not an accident of storage.
Recall
Two ways to get an answer back, depending on where you are:
- Ask directly — "what did I do on this yesterday," "summarize today's standup," "why did we remove that config flag" — and Martium answers by reasoning over adjacent signals, even when there's no single document that spells it out.
- Inline, as you type — a lightweight assistant surfaces suggestions while you write (a Grammarly-style ghost-text pattern, not a coding autocomplete). Low-confidence suggestions stay out of your way entirely; only answers Martium is genuinely confident about get inserted inline, and everything in between shows up in a side panel you can glance at or ignore.
Every answer degrades gracefully: if Martium isn't sure, it says so and shows its sources, rather than guessing with confidence.
Three sources, one answer
Martium blends your personal activity, your team's connected tools, and live search by default — or you can point it at exactly one.
Your activity
Everything Martium has personally watched you do — your commits, your tickets, your debugging sessions, your meetings. This is your memory, private by default.
- Slash command:
/deimos your question - Example:
/deimos what was I debugging in checkout.py last Tuesday
Company knowledge
Whatever your team's GitHub, Jira, Slack, and Confluence connectors bring in — kept current by a background poller that watches for changes, not a one-time import.
- Slash command:
/phobos your question - Example:
/phobos what's the current status of the billing migration
Live search
Pulled in automatically the moment a question needs something current that neither your activity nor your team's tools would know — a new library release, a changed API, current events.
- Automatic when a question needs it
- Or switch to Web mode to search only the open web
By default, Martium decides which source (or blend of sources) actually answers your question — you don't have to know in advance whether the answer lives in your own history, your team's docs, or the open web. The slash commands exist for when you already know exactly where to look.
A context-free AI assistant
Every other AI tool asks you to feed it context before it can help. Martium doesn't need to be briefed — it was already there.
- You brief it first.
- Paste your code. Summarize the ticket. Explain the error.
- Explain what you were doing — again, in a new tab, in a fresh session.
- Context resets the moment you close the window.
- It was already there.
- It watched you debug it, wire it up, read the docs.
- You don't brief it. You just ask.
- Context persists across days, tools, and tasks — because it was captured as it happened, not pasted after the fact.
"No context window. No copy-paste. No re-explaining yourself to a tool that forgot you existed."
For a team, this compounds: the fifth person to ask "why is this built this way" gets the same instant, grounded answer as the first — nobody has to be the designated historian.
A knowledge-transfer layer
Because Martium already saw the work happen, onboarding a new teammate or handing off a project no longer means writing docs from memory after the fact.
New hire onboarding
A new engineer can ask "how does auth work here" or "what's the deploy process" and get an answer grounded in the actual history of decisions — not a stale onboarding doc from eighteen months ago.
Handoff without the handoff meeting
When ownership of a project changes hands, the incoming owner can query the full debugging and decision history directly, instead of scheduling an hour with the person who's already moved on.
Institutional memory survives departures
Reasoning and context that used to leave with an employee stays with the team — the "why," not just the "what," was captured continuously rather than reconstructed from memory during exit interviews.
See it on your own workflow
Early access is onboarding testers in small batches. Reserve a spot and we'll reach out.