You saved a dozen memories yesterday. Today you need to find them. You try searching, but you get a flat list of results with no sense of which memories belong together or what they were about.

You know you worked on "that API thing" in the morning. But your search returns memories from three different sessions, all mixed together. Which ones were from the same work session? Which conversation fixed the actual bug versus which one was just exploring?

This is the context problem. Individual memories are useful. But memories grouped by work session—by conversation—are infinitely more valuable.

The Missing Link: Conversations

When you work on something, you don't create isolated memories. You have a conversation about a topic:

You start investigating a bug
You try several approaches
You find the root cause
You implement the fix
You verify it works

These memories belong together. They tell a story. Searching for them individually loses that narrative thread.

Until now, MemNexus tracked memories but not the conversations they came from. You could search memory content, but you couldn't answer questions like:

What conversations happened today?
What work sessions did I have this week?
Show me all memories from that debugging session Tuesday afternoon

Two new features fix this.

Feature 1: Recent Conversations

See what you actually worked on:

mx conversations list --recent 24h

Output:

Recent Conversations

conv_abc123 - 6 memories - 2h ago
Payment bug investigation
Last: "Fixed race condition with proper mutex"

conv_def456 - 3 memories - 5h ago
Auth system refactor
Last: "Migrated to JWT, all tests passing"

conv_ghi789 - 2 memories - 7h ago
Database query optimization
Last: "Added composite index, 10x faster"

3 conversations - 11 total memories

One command shows you:

Every work session from the last 24 hours
How many memories in each
The last thing you saved (usually the outcome)
When it happened

Time Windows

Filter by any timeframe:

# Today's work
mx conversations list --recent 24h

# This week
mx conversations list --recent 7d

# Last hour (quick status check)
mx conversations list --recent 1h

# Last two weeks
mx conversations list --recent 2w

The format is simple: a number followed by h (hours), d (days), or w (weeks).

Why This Matters

Monday morning standup:

mx conversations list --recent 3d

Instantly see every work session from Friday and the weekend. You know exactly what you completed, what's in progress, and what blocked you.

Context switching: You worked on three different things today. Which one was almost done? Which one needs more investigation?

mx conversations list --recent 24h

Each conversation shows its status via the last memory. "Fixed and deployed" versus "Still investigating" becomes immediately clear.

Weekly review: What did you actually work on this week?

mx conversations list --recent 7d

Not a list of individual tasks—a list of complete work sessions with outcomes.

Feature 2: Search Results by Conversation

The other problem: search results are a flat list. Even if you're searching for related work, results from different sessions get interleaved.

mx memories search --query "API improvements" --group-by conversation

Output:

Search Results: "API improvements"
Grouped by conversation

Conversation: API rate limiting (conv_abc123)
3 memories - 2026-02-03

mem_001 (0.89) - 14:23
Started implementing rate limiting for public API endpoints

mem_002 (0.87) - 15:42
Added Redis-based rate limiter, 100 req/min default per user

mem_003 (0.85) - 16:15
Rate limiting deployed. All tests passing, monitoring looks good


Conversation: API authentication refactor (conv_def456)
2 memories - 2026-02-01

mem_004 (0.81) - 09:30
Switching API auth from API keys to JWT tokens

mem_005 (0.79) - 11:45
JWT auth complete. Added refresh token rotation for security

5 results in 2 conversations

Instead of five disconnected results, you get two complete work sessions. Each shows:

All related memories together
When the session happened
The full narrative from start to finish
Relevance scores for individual memories

Understanding the Results

Memories stay in context: You see that the rate limiting work was a three-step process on the same day. The auth refactor was separate work, different day, different conversation.

Outcomes are visible: The last memory in each conversation usually shows the outcome. "Deployed and monitoring" versus "Still in progress" tells you which work is done.

Timeline is preserved: Memories within each conversation appear in chronological order. You see how the work unfolded, not just what was mentioned.

When to Use This

Finding related work:

mx memories search --query "checkout flow" --group-by conversation

See every work session about the checkout flow. Maybe you fixed three different bugs in three different sessions. Grouped results show that clearly.

Understanding decisions:

mx memories search --query "database choice" --group-by conversation

Decisions often span multiple memories in a conversation: alternatives considered, trade-offs discussed, final choice made. Grouping shows the full discussion, not just fragments.

Debugging archaeology:

mx memories search --query "payment failures" --group-by conversation

When did you work on this? Was it one investigation or multiple attempts? Grouped results show distinct debugging sessions with their outcomes.

Combine Both Features

The real power comes from using these together.

Start broad:

mx conversations list --recent 7d

See all your work sessions from the week.

Drill into relevant ones: You notice "API improvements" came up in multiple sessions. What exactly did you do?

mx memories search --query "API improvements" --group-by conversation

Now you see each improvement session separately, with full context.

Get specific: One conversation looks relevant. Get all its details:

mx memories get conv_abc123

Real Scenarios

Resuming After a Break

You took three days off. What were you working on before you left?

mx conversations list --recent 7d

You see four conversations. Three are completed ("Deployed to production"). One shows "Started investigation, needs more testing." That's where you pick up.

Handoff to Teammate

Your teammate is taking over a project. What work has been done?

mx memories search --query "user preferences feature" --group-by conversation

They get a chronological view of every work session: initial planning, first implementation attempt, the pivot after code review, and the final working version. Complete context, minimal explanation needed.

Working With AI Assistants

When an AI assistant has access to MemNexus, conversation grouping becomes crucial.

Without grouping: "What did I decide about authentication?" The AI sees five memories from two different sessions with different conclusions. It hedges or picks the wrong one.

With grouping: The AI understands there were two conversations. It can see that the second conversation superseded the first. It confidently answers with the current decision and mentions the previous approach was explored but abandoned.

Weekly Review

What did you accomplish this week?

mx conversations list --recent 7d

Instead of a firehose of individual memories, you see 12 complete work sessions. Each has a clear start and outcome. You can quickly identify:

What got finished
What's in progress
What needs follow-up

The Difference

| Before | After | |--------|-------| | Flat list of memories | Work sessions with context | | "What did I save yesterday?" | "What did I work on yesterday?" | | Search returns fragments | Search returns complete stories | | Manual reconstruction needed | Context automatically preserved | | Hard to resume work | Clear continuation points |

Getting Started

Available in CLI v1.7.25:

mx update
mx --version  # Should show 1.7.25

See recent conversations:

mx conversations list --recent 24h

Search with conversation grouping:

mx memories search --query "your project" --group-by conversation

Why Conversations Matter

Memory isn't just about storing information. It's about preserving context.

A single memory answers "what did I learn?" A conversation answers "what was I doing?"

When you come back to work after a break, you don't need individual facts. You need to understand where you were in the story. Which problem were you solving? How far did you get? What was the outcome?

Conversation-based retrieval gives you that. Not just memories—the context they came from.

Conversation-based memory retrieval is available now in MemNexus CLI v1.7.25. Update with mx update or install with npm install -g @memnexus-ai/cli.

Find What You Actually Worked On: Conversation-Based Memory Retrieval