Persistent memory that survives every AI coding session
Jolli Memory remembers what your repo can't: the reasoning behind every commit, portable across every agent you use. It runs entirely on your machine. Your memories are private by default. Stop re-explaining your codebase, ship more, debug less.
curl -fsSL https://jolli.ai/install.sh | sh$
The problem
Your reasoning dies at commit
The reasoning behind every AI coding decision lives in the conversation, and the moment you commit, it's gone.
The why
The part worth keeping
The “why” is everything the diff can't show:
- Why this approach was chosen
- Why the alternatives were rejected
- Why the roadblocks appeared
- Why these tools and agents were used
- Why the context mattered
Why it matters
Context that compounds
Keep the reasoning and every session builds on the last:
- Resume any session without re-explaining yourself
- Hand off work with the reasoning intact
- Review pull requests with full context
- Faster debugging with historical team context
- Cut RCA to minutes with the captured reasoning
How it works
Capture context, reuse it anywhere, and manage it as a living knowledge base and workflows.
Capture
You already run agent skills to plan, implement, commit, and open PRs. Jolli captures the reasoning and context behind each into structured memory, per commit and rolled up per branch.
>
Reuse
Any agent recalls the exact reasoning and context it needs, then reuses those same skills to plan, code, review, and debug, with no re-explaining.
>
Manage
Build workflows on your memory: living specs, doc sites, and token, skills, and risk reports, run on a schedule.
Team standup
Time-basedActive · Last run 6h ago · Succeeded
Daily at 9:00 AM (PDT)
jolliai/jolli / main
Summarize the last 24h of activity into a standup-style status doc…
Company space
Local-first: it works privately on your machine, and syncing to the cloud is opt-in.
Platform agnostic
28+ tools you already use
The surface you build on
Every coding agent Jolli registers into.
The tools you already call
Any MCP, captured as reusable context.
The repos you already host on
Every major Git host, out of the box.
Built to keep your agent fast
What Jolli Memory is engineered to deliver.
Fewer tokens burned
Exact context, not bulk file dumps, with per-commit token accounting.
Faster agent performance
Agents find answers in the memories instead of re-reading the repo.
Captured by default
No stale docs. Memory updates as you commit.
“Jolli turns what you learn working with LLMs into reusable context, closing the loop on context engineering.”
Use cases
Whether you ship solo or lead a team, Jolli Memory turns everyday coding into reusable context.
Individual developer
Keep contextual memory of your coding sessions to cut time spent and tokens burned.
- Speed up PR reviews and debugging with context
- Recall and resume coding sessions instantly
- Hand off a branch with full context intact
- Auto-generate specs, docs, and risk reports
- Plan features against your knowledge graph
Developer manager
Keep team contextual memory to understand how your team performs.
- See token and cost usage across the team
- See which agents, tools, and workflows work best
- Build RCAs fast with full context
- Surface compliance and security risks early
- Keep internal and external docs up to date
Your context shouldn't die at the end of a session.
Stop re-explaining your codebase
Start building on it
curl -fsSL https://jolli.ai/install.sh | shWhat is Jolli Memory?
A local-first memory layer for your codebase. It captures the why behind every commit (the reasoning, alternatives, tokens, and agents) and rolls it up per branch into a browsable, searchable memory bank you and your agents can build from.
How do I start using Jolli Memory?
Three steps, mirroring how it works.
- Capture: install the CLI from npm, add the Claude Code plugin from our marketplace, or use the bundled VS Code / IntelliJ extension — any of them captures your AI coding conversations and summarizes them into context automatically.
- Reuse: a bundled local MCP server (personal context) and cloud MCP server (shared team context) let any agent query, recall, plan, and code against your memory.
- Manage: Jolli Web auto-syncs memories into a browsable Wiki and graph, plus workflows for token and skills reports, drift and gap/risk analysis, and always-current specs and doc sites.
How is it different from agent memory files (like your claude.md)?
Those are static notes you write and maintain by hand: semantic memory that goes stale. Jolli auto-captures the episodic why/what/how on every commit, version-controlled and portable across the agents you use. It can even suggest improvements to your memory and skills files, instead of leaving them to you.
How is it different from AI code review tools?
Code review inspects a diff for issues at pull-request time. Jolli captures the reasoning and context behind changes continuously, as you commit, and makes it reusable later. They're complementary, not competing.
How is it different from inference or token optimizers?
Those speed things up at the model layer. Jolli optimizes the context layer, feeding agents the exact memory they need so they burn fewer tokens and answer faster. It's model-agnostic and sits above any provider you use.
How does it capture context?
On every git commit, Jolli analyzes your changes and writes structured memory (the reasoning, linked files, and decisions behind them) per commit, rolled up per branch into a local knowledge base. No manual note-taking; your memory stays current as you work.
Which editors and agents does it work with?
Jolli Memory works with CLI agents like Claude Code, Codex, Gemini CLI, and OpenCode — the CLI is our recommended way to run it and all you need to get started. In Claude Code you can install it as a plugin from our marketplace instead, which bundles the CLI as an MCP server plus session hooks. Prefer to stay in your editor? It also ships as an extension for VS Code, Cursor, Windsurf, and IntelliJ. Either way, your context stays portable across the tools, MCPs, and Git hosts you use.
Is my memory private?
Yes. Your memories are stored locally on your machine and are yours by default. The only time anything goes out is when it's sent to the model you chose during setup — and syncing to the cloud or sharing a Space with your team is always opt-in.
Do I need an account to start?
No. Download the extension or CLI and start capturing memory for free, no registration required. Add Team Cloud later for shared Spaces, hosted Sites, API access, and team token, cost, and risk reports.
Still have questions? Book a demo