Skip to main content

Saving & Resuming Conversations

The Power of Conversation Checkpointing​

In Lesson 3, you learned the /chat family of commands. Now we dive deep into how to use them strategically to maintain long-running projects.

Key insight: You have two persistence systems:

  • GEMINI.md (Lesson 4) = What Gemini should know about your PROJECT
  • Saved conversations (this lesson) = What you've DISCUSSED with Gemini

Together, they create complete context persistence.

Understanding Conversation Checkpoints​

What Gets Saved?​

When you run /chat save my-project, Gemini CLI saves:

✅ Conversation history — Everything you and Gemini discussed ✅ Context built up — Knowledge shared across the conversation ✅ Decisions made — What you decided on together ✅ Code reviewed — Files you examined ✅ Current state — Where you left off in the work

❌ NOT saved: Your actual code files (those stay in Git)

The /chat Command Family​

CommandPurpose
/chat save <tag>Save current conversation
/chat resume <tag>Restore a saved conversation
/chat listShow all saved conversations
/chat delete <tag>Remove a saved conversation
/chat share file.mdExport conversation to markdown

Naming Your Saved Conversations​

Good names help you find checkpoints quickly:

✅ Good names:

  • auth-jwt-complete
  • database-schema-v2-approved
  • email-feature-started
  • debugging-nov-7-morning

❌ Avoid:

  • save1, save2, checkpoint (too generic)
  • test, temp (confusing)

Naming pattern: [feature]-[status] or [date]-[task]

Conversation Lifecycle: Save → Resume → Update​

The Cycle​

Step 1: Work on something

You: Help me implement JWT authentication
Gemini: I'll help. Here's the approach...
(You discuss and work for 2-3 hours)

Step 2: Save when done with a meaningful chunk

You: /chat save jwt-auth-complete
Gemini: ✓ Saved conversation as "jwt-auth-complete"

Step 3: Resume next time you want to continue

(Later, in new session)
You: /chat resume jwt-auth-complete
Gemini: ✓ Resuming conversation "jwt-auth-complete"
        (Shows conversation history)

You: Now let's add refresh token rotation

Step 4: Continue your work seamlessly

Everything continues from where you left off. No re-explaining.

Real Workflow: Multi-Day Project​

Day 1: Authentication Foundation​

$ gemini

You: Let's build JWT authentication. Here are the requirements:
     - Register endpoint
     - Login endpoint
     - Protected routes with JWT validation
     - Refresh tokens

Gemini: I'll help design this...
(Spend 2 hours designing and discussing)

You: /chat save auth-endpoints-designed
Gemini: ✓ Saved conversation as "auth-endpoints-designed"

You: /quit

Day 2: Implementation​

$ gemini

You: /chat resume auth-endpoints-designed
Gemini: ✓ Resumed conversation "auth-endpoints-designed"
        (Shows your design discussion from yesterday)

You: Great, let's implement the register endpoint first
(Continue seamlessly from yesterday - Gemini has context)

(After 2 hours)
You: /chat save register-endpoint-complete
Gemini: ✓ Saved

Day 3: Testing & Refinement​

$ gemini

You: /chat resume register-endpoint-complete
Gemini: ✓ Resumed

You: Now let's write comprehensive tests for the register endpoint

(After testing work)
You: /chat save jwt-auth-system-complete
Gemini: ✓ Saved

Next Project: Switch Contexts​

$ gemini

You: /chat list
Gemini CLI:
  - auth-endpoints-designed
  - register-endpoint-complete
  - jwt-auth-system-complete
  - (other projects)

You: /chat save current-work
You: /clear
(New conversation starts with fresh context)

You: /chat resume frontend-redesign-v2
Gemini: ✓ Resumed previous frontend project

Managing Your Saved Conversations​

Listing Conversations​

You: /chat list
Gemini CLI:
  - auth-jwt-complete (Nov 7, 2h 45m)
  - database-migration-in-progress (Nov 5, 1h 20m)
  - api-error-fixes (Oct 28, 45m)
  - frontend-redesign-v2 (Oct 15, 3h 10m)

Exporting Conversations​

Share what you've discussed with others:

You: /chat share project-design.md
Gemini: ✓ Exported conversation to project-design.md

You can now share project-design.md with teammates so they see your design decisions.

Deleting Old Conversations​

Clean up completed projects:

You: /chat delete old-feature-spike
Gemini: ✓ Deleted "old-feature-spike"

Advanced Patterns​

Pattern 1: Experimentation With Safety​

Try a risky approach without losing current work:

You: /chat save current-stable-state
Gemini: ✓ Saved

(Try experimental refactoring)
You: Hmm, this approach isn't working well...

You: /clear
Gemini: ✓ Conversation cleared

You: /chat resume current-stable-state
Gemini: ✓ Loaded "current-stable-state"
(You're back to a known good state)

Pattern 2: Parallel Exploration​

Work on two approaches simultaneously:

You: /chat save approach-a-v1
Gemini: ✓ Saved

You: /clear
Gemini: ✓ Fresh conversation

You: Let me try a different architecture approach...
(Work for 1 hour)

You: /chat save approach-b-v1
Gemini: ✓ Saved

You: /clear
You: /chat resume approach-a-v1
(Compare both approaches later)

Pattern 3: Pair Programming handoff​

Developer A saves their work, Developer B resumes:

# Developer A (morning)
You: [Work on feature for 2 hours]
You: /chat save feature-auth-state-for-bob
You: /quit

# Developer B (afternoon)
You: /chat resume feature-auth-state-for-bob
Gemini: ✓ Resumed Bob's conversation
(Bob's context and decisions are preserved)

Pattern 4: Multi-Hour Sessions with Checkpoints​

Break up long work with periodic saves:

(Work for 45 min)
You: /chat save auth-endpoints-skeleton
Gemini: ✓ Saved

(Work another 45 min on auth tests)
You: /chat save auth-tests-complete
Gemini: ✓ Saved

(Work another 45 min on rate limiting)
You: /chat save auth-rate-limiting-complete
Gemini: ✓ Saved

(Later, if something breaks)
You: /clear
You: /chat resume auth-tests-complete
(Jump back to a known good state)

GEMINI.md + Saved Conversations Together​

These two systems complement each other:

GEMINI.md (persistent, project-wide):

  • What Gemini should know about your project
  • Architecture, decisions, conventions
  • Loaded automatically every session

Saved conversations (per-person, per-task):

  • What YOU discussed with Gemini
  • Your specific design decisions THIS session
  • Context for THIS task continuation

Complete Workflow​

Day 1:
  $ cd auth-service && gemini
  (GEMINI.md loads automatically)
  You: [Discuss architecture for 2 hours]
  You: /chat save arch-design-approved
  You: /quit

Day 2:
  $ cd auth-service && gemini
  (GEMINI.md loads automatically - same as before)
  You: /chat resume arch-design-approved
  (Now you have BOTH the project context AND yesterday's work)
  You: Let's implement the routes now

Day 3 (switch projects):
  $ cd frontend && gemini
  (Different GEMINI.md loads automatically)
  You: /clear
  You: /chat resume frontend-components-v2
  (Now working on frontend with its own GEMINI.md)

Result: You always have the right context for the right project.

Common Questions​

Q: What's the difference between saving conversations and GEMINI.md?

A:

  • GEMINI.md = "Project architecture is X, we use TypeScript, our conventions are Y"
  • Saved conversation = "We discussed X yesterday, decided on approach Y, need to implement Z today"

Save conversations for work continuity. Use GEMINI.md for project knowledge.

Q: Can I resume someone else's conversation?

A: Not directly—saved conversations are personal. But you can export with /chat share and send to a teammate.

Q: If I don't save, is my conversation lost?

A: Yes. Always save important work with /chat save before closing Gemini CLI.

Q: Can I rename a saved conversation?

A: Not directly. Use /chat export to file, then save as new conversation.

Q: How long are conversations kept?

A: Conversations are stored locally on your machine. They persist until you delete them with /chat delete.

Q: How much storage do saved conversations use?

A: Conversations are text, so very little. A typical 2-hour conversation might be 500KB-2MB.

Exercises​

Exercise 1: Save Your First Conversation​

Have a real conversation about something meaningful:

$ gemini
You: [Work on a real task for 20+ minutes]
You: /chat save first-saved-session
Gemini: ✓ Saved

Expected: You can now resume this conversation anytime.

Exercise 2: List and Inspect​

You: /chat list
(See all your saved conversations)

You: /chat save another-session
You: /chat list
(See it appears in the list)

Exercise 3: Resume and Continue​

Close Gemini CLI. Reopen it:

$ gemini
You: /chat resume first-saved-session

Expected: Your conversation history is restored. Gemini remembers what you discussed.

Exercise 4: Export and Share​

You: /chat share my-design.md
(Check that my-design.md was created)

You: cat my-design.md
(Read the exported conversation)

Exercise 5: Multi-Session Workflow​

Save, clear, resume:

You: [Work for 10 minutes]
You: /chat save session-one
You: /clear
(Fresh conversation - old context is gone)

You: [Work on something different for 10 minutes]
You: /chat save session-two
You: /clear

You: /chat resume session-one
(Conversation one context restored)

Best Practices​

✅ DO​

  • ✅ Save when you complete meaningful work
  • ✅ Use clear, descriptive names
  • ✅ Save before switching projects
  • ✅ Export important decisions with /chat share
  • ✅ List your conversations occasionally to stay organized
  • ✅ Delete old completed projects to keep list clean

❌ DON'T​

  • ❌ Forget to save—close Gemini without /chat save
  • ❌ Use generic names like "save1", "temp", "test"
  • ❌ Accumulate dozens of conversations (clean up periodically)
  • ❌ Assume conversations are synced across devices (they're local)

Key Takeaways​

  • Saved conversations persist your work across sessions
  • /chat save <tag> saves current conversation
  • /chat resume <tag> loads a previous conversation
  • /chat list shows all saved conversations
  • /chat share file.md exports for sharing with others
  • Combined with GEMINI.md = complete context persistence
  • Use clear names like feature-complete or date-task
  • Save strategically after completing meaningful chunks
  • Clean up periodically with /chat delete

Next lesson: You'll learn MCP servers to extend Gemini CLI with external integrations (web browsing, API documentation, GitHub).