cd /news/ai-agents/track-neon-sessions-and-keep-context… · home › topics › ai-agents › article
[ARTICLE · art-172692] src=entire.com ↗ pub= topic=ai-agents verified=true sentiment=↑ positive

Track Neon Sessions and Keep Context Across Models with Entire

Entire released a walkthrough showing how its Entire CLI and entire-brain tooling capture agent sessions as checkpoints and hand context between models, demonstrated by running Gemini's Neon model in Antigravity CLI and then switching to Gemini Carbon. In the author's run, entire brain brief --handoff returned only a session heading, so Carbon recovered the watch-party planner rules from the saved checkpoint and project files instead of from retrieval. The piece recommends checking what a new model actually retrieved before letting it continue coding.

recommendation

Checkpoint agent sessions with Entire and inspect the handoff brief before switching models; if retrieval returns only a heading, have the new model read the saved checkpoint and files instead.

more in playbook: Designing agent systems
by read3 min views3 publishedNov 19, 2026
Track Neon Sessions and Keep Context Across Models with Entire
Image: Entire (auto-discovered)

I switch models a lot and then a week later, I’m looking at a diff thinking, “Why did my agent do that?” Entire saves the conversation behind the change, and then entire-brain gives the next agent a way to look it up.

This week, I’ve been loving Gemini’s new Neon model. So let’s try it out on a Warriors watch party food planner.

Before You Start #

You’ll need Antigravity CLI with Neon access, Entire CLI, Python 3, and a Git repo with at least one commit before you launch the agent. Run the commands from that repo’s root.

Connect Entire and Launch Neon #

Enable tracking, then launch Neon from the repository root:

$entire enable --agent antigravity$agy --model gemini-4-neon-high

If Entire is already enabled, replace the enable command with .

Let Neon Handle the Snack Math #

Paste into Antigravity:

Create watch_party.py with plan_watch_party(guests, vegetarians). Return quantities for 4 veggie pinwheels and 2 fruit skewers per guest, plus 2 chicken sliders per non-vegetarian guest. Keep zero quantities. Use finger foods without peanuts. Require guests > 0 and 0 <= vegetarians <= guests; reject non-integers and booleans. Add test_watch_party.py using unittest; test mixed and all-vegetarian groups plus invalid inputs. Run the tests. Don’t commit yet.

Check Capture, Then Commit #

$entire session list

Run this command in another terminal in the same repo. If your session is missing, check the first commit and restart Antigravity after hook setup.

Read the files and test results, then ask your agent to commit only watch_party.py and test_watch_party.py. Now peek at the conversation with:

$entire checkpoint explain --commit HEAD --full

You should see a checkpoint, your prompt, and the response. Mine had the snack math and the passing test results!

Switch Models, Keep the Context #

Since the planner worked so well, I decided to add a grocery checklist and switch models without repeating the food rules.

To try the handoff, commit any follow-up work you want captured, then review and commit the generated Entire config. From a clean worktree, run the first command if Brain isn’t installed, then start Brain:

$curl -fsSL https://raw.githubusercontent.com/entireio/entire-brain/main/scripts/get-brain.sh | bash$entire brain

Setup can finish with some sources unavailable. Before passing the baton, check what the handoff actually contains:

$entire brain brief --handoff

For checkpoints created after setup, run first. Exit Neon and start a fresh session with another model available to your account–here, I used Carbon.

$agy --model gemini-4-carbon

Then ask:

Use Entire Brain to find the planner session. Summarize its rules with sources before editing. If retrieval is empty, say so and inspect the saved checkpoint and files. Distinguish uncommitted work from checkpointed history.

In my run, Brain returned only a session heading. Carbon recovered the rules from the saved checkpoint and project files instead.

New Model, Old Game Plan #

As you hand off from model to model, check what the new model actually found before asking it to keep coding. If something’s missing, start with for capture or for retrieval.

Try it yourself and tell us what you’re building on our Discord, and Go Warriors!

── more in #ai-agents 4 stories · sorted by recency
── more on @entire 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/track-neon-sessions-…] indexed:0 read:3min 2026-11-19 · —