Auto-capture
Save finished conversations without exporting anything.
Vesti · working memory for agents
The working-memory engine for domain-specific agents.
Vesti captures every agent conversation, organizes it into structured local memory, and serves it back to any agent that needs the context.
Vesti library
Search and reopen
Enterprise · FDE
Enterprise knowledge agents, delivered FDE-style.
Our engineers embed with your team and turn one real scenario into a running, measurable, re-testable agent system in 4–20 weeks — private deployment, evidence chains, humans at the final gate.
Explore Enterprise- 4
- FDE deliveries to date
- 1
- SOE lighthouse delivered
- 60→2
- minutes for a compliance answer
Team
Seven disciplines, one memory stack.
An interdisciplinary team from Nanjing University, Fudan and SJTU — gold medalists at the AI Hackathon Tour national finals and GOSIM Paris 2026 Frontier Creators. Zero to launch in 63 days, 463 commits.
Meet the teamFeatures
Short on ceremony, strong on recall.
Fast search
Find past prompts, notes, and decisions in seconds.
Local-first
Keep your conversation history on your own machine.
Clear recall
Reopen the exact thread instead of starting over.
Demo
See the memory loop in one pass.
Capture the thread, search it later, and reopen the context fast.
Skills
Open-source skills for coding agents.
VESTI Skills (MIT) make your AI coding agent remember everything you have done — and hand the work over to the next agent, intact.
vesti-memory
Memory recall for any agent session.
Progressive disclosure: pull the project context pack at session start, then drill from session index to file locations to exact turns.
- Auto-loads a context pack on session start — state card, active files, open questions.
- Drills down through vesti_search, vesti_search_files, vesti_timeline and vesti_get_turns.
- No more re-briefing a new agent, or losing decision rationale after /compact.
Requires the VESTI desktop app running locally with vesti-mcp registered.
vesti-handoff
Structured handoff between agents.
Generates a schema-based handoff pack — goal, state, files, failedPaths, verification, nextSteps — so the next agent starts from proof, not trust.
- One rule: verify before you take over — rerun verification.lastCommand first.
- Works standalone, no VESTI dependency.
- Schema-aligned with the VESTI desktop app's Relay Pack.
Standalone — works without VESTI.
Install
Add the skills to your agent.
One command in Kimi Code or Claude Code — or copy the skills manually into any agent's skills directory.
GitHub — VESTI-SKILLSJust tell your agent
Install the VESTI skills from GitHub: clone https://github.com/firefly-hefeng/VESTI-SKILLS to a temp dir, copy skills/vesti-memory and skills/vesti-handoff into your user-level skills directory (~/.kimi-code/skills/ for Kimi Code, ~/.claude/skills/ for Claude Code, or your tool's equivalent), then list what you installed and tell me whether a restart or new session is needed.
Paste this to Kimi Code, Claude Code, Codex or any shell-capable agent — it clones, copies and verifies by itself. Skills activate in a new session.
One-command install
RecommendedKimi Code
/plugins install https://github.com/firefly-hefeng/VESTI-SKILLSRun /reload afterwards to activate.
Claude Code
/plugin marketplace add firefly-hefeng/VESTI-SKILLS/plugin install vesti-skills@vesti-skillsManual install
git clone https://github.com/firefly-hefeng/VESTI-SKILLS.git| Agent | User-level | Project-level |
|---|---|---|
| kimi-code | cp -r skills/<name> ~/.kimi-code/skills/ | .kimi-code/skills/ |
| Claude Code | cp -r skills/<name> ~/.claude/skills/ | .claude/skills/ |
| Codex / Cursor / others | Import the full SKILL.md per the tool's skills/prompt convention | Same as user-level |
vesti-handoff works standalone. vesti-memory needs the VESTI desktop app (or @vesti/mcp) providing the MCP tools.
Download
Two products, one memory loop.
Capture agent work where it happens — in the browser, and in your local CLI coding agents.
Browser extension · v1.2.0-rc.8
心迹 Vesti for the browser
Auto-capture conversations across ChatGPT, Claude, Gemini, DeepSeek, Tongyi, Doubao, Kimi and Yuanbao — structured, searchable, local-first.
Desktop app · Windows · v0.3.2
Vesti for coding agents
Capture Claude Code, Cursor, Kimi Code and Codex sessions into an L0–L3 layered memory system, with a built-in MCP server any agent can mount.
Windows may show a blue SmartScreen prompt on first install — click "More info → Run anyway" (the installer is not code-signed yet; a certificate is in the works). Verify the installer SHA-256 below.
SHA-256: 06da5f6363df9772509a0e456788ec43698307f7804220e83a6387d42239d886
The macOS build is coming soon.
Installers are served directly from this site: Windows and a self-hosted Linux AppImage are available now, with macOS on the way.