From experience to better agents

Self Evolve
Your Agent.

Turn past conversations into lessons your agent can use. Review proposed improvements, reuse what works, and keep your goals, results, and feedback connected.

Send one prompt to your agent. Your own Codex. A clearer direction.

✦ LuLuLearning loop · Example

A better playbook, one lesson at a time.

“Keep what worked. Learn from what didn’t.”
01
SELECTED WORK

Conversations & corrections

Start with your recent Codex conversations.

02
LESSONS WITH EVIDENCE

A draft worth reviewing

Keep the source behind every suggestion.

✓
YOUR APPROVAL

Guidance you choose to keep

Review first. Save the changes you trust.

↻
Ready for your next task

Reusable guidance. A clearer direction.

Your agent. Your experience. Your judgement.
Agent ecosystem

Research reference · ICLR 2026

Turn experience into reusable guidance.

ReasoningBank research figure comparing memory formats and cumulative successful tasks
ReasoningBank paper, Figure 1 · Reusable reasoning memory and cumulative success on WebArena-Admin.
ReasoningBank memory retrieval, extraction and consolidation research diagram
ReasoningBank paper, Figure 2 · Retrieve memory, extract lessons, consolidate new experience.

LuLu helps your agent extract lessons from past work, keep the evidence, and carry approved guidance into future tasks.

Published reference benchmarks · Gemini 2.5 Flash · compared with No Memory

WebArena success rate

+8.3 percentage points

40.5% → 48.8%

684 tasks · Table 1

Fewer interaction steps

14.4%

9.7 → 8.3

WebArena average · calculated from Table 1

SWE-Bench Verified resolve rate

+4.6 percentage points

34.2% → 38.8%

Software engineering · Table 2

Source: Ouyang et al., ReasoningBank, ICLR 2026. Read the paper ↗ · Research repository ↗ · Apache 2.0

From conversations to outcomes

Less keeping track. More moving forward.

A goal is more than a task. Keep the purpose, the work, and your feedback connected as your understanding evolves.

01 / ORGANIZE

Learn from real work.

Use recent Codex conversations to uncover goals, retries, and corrections. Review improvements with evidence from the work that produced them.

02 / EXPLORE

Run your goals like a city.

Buildings are goals. Roads connect the work. Explore your city, open a goal, and see exactly where things stand.

03 / GUIDE

You choose what changes.

Read your agent's results, review proposed guidance, and approve the changes you want to keep. Give the next round a clearer direction.

01 / From real work

Good guidance starts with real evidence.

Choose recent Codex conversations. LuLu helps uncover retries, corrections, and useful patterns, with short evidence excerpts you can review.

  • See the work behind a suggestion.
  • Review the lesson in its context.
  • Raw conversation history stays on your computer.
From conversation to lessonIllustrative example
A CORRECTION

“Run the required checks before releasing the change.”

↓
PROPOSED GUIDANCE

Make verification part of the workflow.

State what passed, what was skipped, and what still needs attention.

Evidence attached · Awaiting review

02 / Work you repeat

Build a playbook for the work you repeat.

From a coding session to a research task, keep useful corrections within reach. These examples show the kinds of guidance you can review and reuse.

Connect your own Codex ↗
One playbook. Different work.Examples
01 / DEVELOPMENT

Carry release checks into your next coding task.

✓ Run the checks that match the change.
✓ Verify the published behavior.
✓ Keep the result and its evidence together.
Illustrative guidance · You decide what to keep

03 / You stay in control

Your judgement stays in the loop.

Read the evidence, edit the proposal, and approve what deserves to become guidance. Keep useful changes as a Prompt, then confirm any update to your project instructions.

  • Drafts remain drafts until you approve them.
  • Keep feedback close to goals and results.
  • Choose the guidance for your next round of work.
Review an improvementIllustrative example
Awaiting your review

A clearer release checklist

Use the source evidence to decide which checks belong in your project guidance.

↗ AGENTS.mdProposed change
Read evidenceEdit draft✓ Approve guidance
Review → Save Prompt → Confirm project update

Let your agent get you started

Keep your Chat. Get the bigger picture.

Install LuLu, connect your account, and run your first analysis. Your goals become a city you can explore.

Install →Authorize→Analyze→Explore

macOS / Linux · Codex installed and signed in

Send this prompt to your agent

Please install and connect Lulu on this computer: https://lulucli.com/proactive. If it is already installed, check and restore the local connection. Otherwise install using https://lulucli.com/proactive/install.sh and connect to https://lulucli.com/proactive. Open the sign-in page if needed, then link this computer automatically. Finally verify that this computer is connected to the current account and website.

Or install with a command
curl -fsSL https://lulucli.com/proactive/install.sh | sh
Your work stays yours.

Raw conversation history stays on your computer. Selected excerpts are analyzed by your configured Codex model service. LuLu saves goals, results, and the evidence needed to keep them grounded.

Contact

Building a clearer way to work with AI.

Silicon Valley

cjc@lulucli.com