You ask your AI assistant to reconstruct what happened during yesterday's debugging session. It dutifully searches your memories and returns a list of results—sorted by relevance score. Now you're manually piecing together the timeline, scrolling through results, trying to figure out what happened first, what led to what, and where the breakthrough finally occurred.

This shouldn't be your job.

The Hidden Cost of Relevance Sorting

Memory search has traditionally optimized for one thing: relevance. Given a query, return the most semantically similar results. This works great for finding specific information: "What's our database password policy?" or "How do we deploy to staging?"

But it completely falls apart for a different—and equally common—use case: understanding what happened.

When you're trying to:

Reconstruct a debugging session
Review how a decision evolved
Understand the sequence of events that led to an outcome
Brief someone on yesterday's progress

...relevance scores are actively unhelpful. A crucial early observation might score lower than a later summary. The breakthrough moment might be buried between setup steps and follow-up tasks.

The Five-Search Problem

We ran an experiment. We asked an AI assistant to reconstruct a morning's work on a development project. Here's what happened:

First search: Found recent memories but no temporal context
Second search: Tried different keywords to find earlier work
Third search: Searched for specific topics to fill gaps
Fourth search: Looked for memories with "started" or "beginning"
Fifth search: Finally had enough pieces to attempt reconstruction

Five searches. Manual sorting. Guesswork about temporal ordering. And the result was still incomplete because auto-saved terminal outputs cluttered the results.

The AI spent more time searching than synthesizing.

Introducing Timeline Search

Today we're releasing Timeline Search in MemNexus CLI v1.7.19—a new way to search that's optimized for temporal understanding.

The `--timeline` Flag

mx memories search --query "payment debugging" --timeline

Instead of relevance-sorted results, you get:

Timeline: "payment debugging"

2026-02-04 09:15  [CURRENT]  mem_abc123
Started investigating payment failures. Customer report indicates
intermittent 500 errors on checkout.

2026-02-04 09:42  [CURRENT]  mem_def456
Found correlation with high traffic periods. Hypothesis: connection
pool exhaustion under load.

2026-02-04 10:23  [SUPERSEDED]  mem_ghi789
Tried increasing pool size to 50. No improvement.

2026-02-04 11:15  [CURRENT]  mem_jkl012
Root cause identified: race condition in retry logic. Multiple
threads claiming same connection.

2026-02-04 11:58  [CURRENT]  mem_mno345
Implemented fix with connection-level mutex. All tests passing.

5 memories spanning 2h 43m - 1 superseded

One search. Complete timeline. Clear state indicators.

Timestamps in Default Search

Even without --timeline, search results now include timestamps:

mx memories search --query "deployment"

| Score  | Time          | ID         | Content                         |
|--------|---------------|------------|---------------------------------|
| 0.89   | 02-04 14:23   | mem_abc... | Deployed v2.1.0 to production   |
| 0.84   | 02-03 09:15   | mem_def... | Staging deployment successful   |
| 0.81   | 02-01 16:42   | mem_ghi... | Updated deployment pipeline     |

Now you can see at a glance when things happened, even in relevance-sorted results.

Filtering the Noise

Timeline reconstruction has another enemy: noise. Auto-saved terminal outputs, session snapshots, and routine logs can drown out the meaningful memories.

Topic Filtering

Use --topics to focus on intentional, meaningful memories:

mx memories search --query "authentication" --timeline --topics "implementation,decision,completed"

This returns only memories tagged with those topics—filtering out auto-saved noise while preserving the complete picture of your actual work.

Coming Soon: Exclude Topics

We're also working on --exclude-topics (#180) for even more precise filtering:

# Filter out auto-saves without specifying what you DO want
mx memories search --query "authentication" --timeline --exclude-topics "auto-saved,terminal-output"

Real-World Use Cases

Morning Standup Prep

mx memories search --query "yesterday's work" --timeline --recent 24h

Get a chronological summary of everything you worked on, ready to share with your team.

Debugging Session Review

mx memories search --query "memory leak investigation" --timeline --topics "debugging,solution"

Reconstruct the full arc of an investigation—from initial symptoms through hypotheses to resolution.

Decision Archaeology

mx memories search --query "chose React over Vue" --timeline

See not just the decision, but the context: what alternatives were considered, when the discussion happened, and how the conclusion was reached.

Onboarding Context

mx memories search --query "authentication system" --timeline --topics "architecture,decision"

Give new team members a chronological understanding of how a system evolved, not just what it looks like today.

The Technical Details

Timeline search uses the same powerful hybrid search engine under the hood—combining semantic understanding with keyword matching. The difference is in presentation and sorting:

Chronological ordering: Results sorted by eventTime (or createdAt if not specified)
State indicators: Each memory shows its effective state ([CURRENT], [SUPERSEDED], [CONTRADICTED])
Temporal summary: Footer shows time span and state breakdown
Compact timestamps: Optimized date format (YYYY-MM-DD HH:MM) for scanability

The search itself is just as fast—timeline mode adds no additional latency.

Upgrading

Timeline search is available now in CLI v1.7.19:

mx update
mx --version  # Should show 1.7.19

No API changes required. No configuration needed. Just add --timeline to any search.

What's Next

Timeline search is part of our broader vision for temporal intelligence in AI memory. Coming soon:

Automatic timeline generation: Ask for "my week" and get a structured summary
Timeline diff: Compare what you knew at two different points in time
Narrative threads: Group related memories into named storylines
Temporal search operators: "What did I believe about X before Y happened?"

Because memory isn't just about storing information. It's about understanding how knowledge evolves through time.

Timeline Search is available now in MemNexus CLI v1.7.19. Update with mx update or install with npm install -g @memnexus-ai/cli.

Stop Playing Detective: How Timeline Search Transforms Memory Reconstruction