Time Travel

Every document in Sombra has full version history, powered by an immutable database. Your AI agent can look back to any point in time — no manual snapshots, no "save as v2", no git required.

How it works

Sombra is built on an immutable data store where nothing is ever overwritten in place. Every edit to every document creates a new revision with a precise timestamp. The full history is always there — these tools give your AI agent access to it.

changes_since

What changed?

Returns a structured summary of every revision since a given timestamp: which sections were touched, how many lines changed, and when. Compact enough for your agent to scan without burning tokens.

read_snapshot

What did it say?

Returns the full content of a document as it existed at any point in time. Optionally filter to a single section. Your agent can compare then and now without you having to find the right version.

What your agent can do

Resume where you left off

The resume skill uses changes_since to show exactly what happened since your last session:

### Since last session (3 changes)
- 14:32 — Initial spec written (18 lines)
- 14:35 — Audit & Recovery, Decisions updated
- 14:40 — NP-4 checked off in SPA Architecture

Check what a spec said on any date

"I think we messed up the spec, what did it say on Tuesday?" Your agent calls read_snapshot with a section filter and returns just the relevant part — no need to pull the entire document.

read_snapshot(id, kind: "context", as_of: "2026-04-08", section: "## Decisions")
## Decisions
- Use single-blocks-array model
- TipTap for rich text editing
- Server-side snap calculations

Trace how a decision evolved

Filter changes_since to a specific section and see every revision that touched it. Your agent can reconstruct the timeline of a decision without reading every version of the full document.

Designed for agents

Structured summaries, not diffs

changes_since returns section names and line counts — not raw unified diffs. Your agent gets the signal it needs in a fraction of the tokens.

Section filtering

Both tools accept an optional section filter. Ask about "## Decisions" and get just that section — not the entire 30k-token document.

Newest-first ordering

Results come back with the most recent changes first. Your agent usually cares about what just happened, not the initial write.

Works everywhere

Available to any MCP client — Claude Code, Claude Desktop, ChatGPT, OpenCode. No git required.

These tools are part of the Sombra MCP server

Connect your AI agent and start using changes_since and read_snapshot right away.