/AI4h ago

Kocoro Launches Open-Source AI Agent With Persistent Memory For Mac

11214236.9K
Rohan Paul#1031
Original post
Rohan Paul@rohanpaul_ai#1031inAI

A new open-source agent engine is trying to make AI sessions continuous across days.

🧠 Kocoro is a Mac-native AI agent trying to fix one of the most annoying problems in daily AI work: every new session forgets what happened before.

Kocoro reviews your workday locally every night, then trains a light memory model over the cloud (by using TensorLogic), downloads it back, and the next morning picks up exactly where you left off — no re-pasting, no re-explaining.

It knows where your projects live, what you were changing yesterday, which tools you reach for. It can open the browser, operate desktop apps, organize files, update documents, and carry a task across several apps without you stitching the steps together by hand.

And what's cool is that they have made Kocoro's kernel is open source with a Github repo. Because an AI that operates your computer has to be verifiable, auditable, and under your control. Something that drives your machine shouldn't get there on "trust us."

You can run the kernel from the command line with Shannon, or you can install the desktop app and never touch a terminal. Either way, the process is meant to be open — visible enough to inspect, editable enough to bend, and easy enough to switch off.

Real memory isn't keeping every word. It's knowing what matters, what will shape the next decision, and what changes over time. You don't remember every sentence your colleague said last quarter — you remember that they hate surprise meetings, that the Q3 launch slipped

9:52 AM · Jun 9, 2026 · 3K Views
Sentiment

Users appreciate Kocoro's open-source Mac AI agent because its persistent memory solves the practical issue of retaining useful context without clutter.

Pos
100.0%
Neg
0.0%
1 comments with sentiment.
Cluster Engagement
Posts from X
Most Activity
Most Activity
VIEWS1.5KBOOKMARKS3LIKES5RETWEETS1REPLIES2
Rohan Paul@rohanpaul_ai

Kocoro treats AI memory as a nightly cleanup job: after you finish working, it reads the day’s chats, file changes, and app activity signals.

It extracts the things that actually matter later, such as projects, people, decisions, preferences, and open tasks, instead of saving random keywords.

Those pieces are compressed into a typed knowledge graph model (trains using TensorLogic and downloaded back to your Mac), where every memory has a clear role and every connection has a clear meaning.

Kocoro then prunes weak memories, so the graph model stays small, relevant, and high-signal instead of becoming another messy archive.

The model stays on your Mac, gets queried locally with zero extra LLM calls at runtime, and the next session starts with the right context already loaded before you type.

Rohan Paul@rohanpaul_ai

A new open-source agent engine is trying to make AI sessions continuous across days.

🧠 Kocoro is a Mac-native AI agent trying to fix one of the most annoying problems in daily AI work: every new session forgets what happened before.

Kocoro reviews your workday locally every night, then trains a light memory model over the cloud (by using TensorLogic), downloads it back, and the next morning picks up exactly where you left off — no re-pasting, no re-explaining.

It knows where your projects live, what you were changing yesterday, which tools you reach for. It can open the browser, operate desktop apps, organize files, update documents, and carry a task across several apps without you stitching the steps together by hand.

And what's cool is that they have made Kocoro's kernel is open source with a Github repo. Because an AI that operates your computer has to be verifiable, auditable, and under your control. Something that drives your machine shouldn't get there on "trust us."

You can run the kernel from the command line with Shannon, or you can install the desktop app and never touch a terminal. Either way, the process is meant to be open — visible enough to inspect, editable enough to bend, and easy enough to switch off.

Real memory isn't keeping every word. It's knowing what matters, what will shape the next decision, and what changes over time. You don't remember every sentence your colleague said last quarter — you remember that they hate surprise meetings, that the Q3 launch slipped

4hViews 1.5KLikes 5Bookmarks 3
Rohan Paul@rohanpaul_ai

Their Github repo is open-source, which contains the Kocoro engine + daemon: the shan runtime that does the actual work (agent loop, local tools, permission engine, channel messaging, MCP, scheduling).

It's fully usable on its own via the CLI, TUI, daemon HTTP API, and MCP. Kocoro Desktop — the native GUI app shown above — is a separate, closed-source product that runs on top of this daemon.

Rohan Paul@rohanpaul_ai

Kocoro also prunes the graph, which means it throws away low-value memory instead of saving everything forever.

That is a big design choice because raw chat logs get noisy fast, while a small graph can stay focused enough to be useful at the start of the next session.

4hViews 704Likes 0Bookmarks 2
Rohan Paul@rohanpaul_ai

Download Kocoro Desktop (macOS) — DMG, the recommended way to use Kocoro

CLI only — npm install -g @kocoro/kocoro (build-from-source and other options under Installation) - https://github.com/Kocoro-lab/Kocoro#get-kocoro

🔧 Kocoro daemon — the macOS agent loop → http://github.com/Kocoro-lab/Kocoro

⚙️ Shannon — the backend AI runtime engine

→ http://github.com/Kocoro-lab/Shannon

Scheduling, Chrome CDP, Accessibility API, local file I/O — all in the open.

Follow the creator: @waylandzhang

Rohan Paul@rohanpaul_ai

Their Github repo is open-source, which contains the Kocoro engine + daemon: the shan runtime that does the actual work (agent loop, local tools, permission engine, channel messaging, MCP, scheduling).

It's fully usable on its own via the CLI, TUI, daemon HTTP API, and MCP. Kocoro Desktop — the native GUI app shown above — is a separate, closed-source product that runs on top of this daemon.

4hViews 662Likes 0Bookmarks 2
Rohan Paul@rohanpaul_ai

And because it remembers, it can keep working when you're not watching.

You can schedule it to draft tomorrow's posts, pull a weekly report, summarize the morning's news, and push the result to you in Slack or LINE — so the first thing you see isn't a blank prompt, but work already done.

Rohan Paul@rohanpaul_ai

Kocoro treats AI memory as a nightly cleanup job: after you finish working, it reads the day’s chats, file changes, and app activity signals.

It extracts the things that actually matter later, such as projects, people, decisions, preferences, and open tasks, instead of saving random keywords.

Those pieces are compressed into a typed knowledge graph model (trains using TensorLogic and downloaded back to your Mac), where every memory has a clear role and every connection has a clear meaning.

Kocoro then prunes weak memories, so the graph model stays small, relevant, and high-signal instead of becoming another messy archive.

The model stays on your Mac, gets queried locally with zero extra LLM calls at runtime, and the next session starts with the right context already loaded before you type.

4hViews 904Likes 2Bookmarks 0
Rohan Paul@rohanpaul_ai

Most AI tools handle context in 2 rough ways: they either keep a longer chat window, or they retrieve old text chunks using RAG, which means retrieval-augmented generation.

Kocoro is different as it does not want to replay old text, it wants to remember the meaning of what happened.

Rohan Paul@rohanpaul_ai

And because it remembers, it can keep working when you're not watching.

You can schedule it to draft tomorrow's posts, pull a weekly report, summarize the morning's news, and push the result to you in Slack or LINE — so the first thing you see isn't a blank prompt, but work already done.

4hViews 28Likes 0Bookmarks 0
Rohan Paul@rohanpaul_ai

Kocoro also prunes the graph, which means it throws away low-value memory instead of saving everything forever.

That is a big design choice because raw chat logs get noisy fast, while a small graph can stay focused enough to be useful at the start of the next session.

Rohan Paul@rohanpaul_ai

The super useful part is the graph structure.

A graph means Kocoro does not only store “task X exists,” but can connect it to a project, a person, a decision, and a next action, so the agent can recover the working context without asking the user to paste everything again.

4hViews 28Likes 0Bookmarks 0
Rohan Paul@rohanpaul_ai

Kocoro works like Mac-native AI coworker: you type normal requests, and it can read files, edit code, schedule reminders, connect to tools like GitHub or Slack through MCP, and run background agents for recurring work.

Rohan Paul@rohanpaul_ai

Most AI tools handle context in 2 rough ways: they either keep a longer chat window, or they retrieve old text chunks using RAG, which means retrieval-augmented generation.

Kocoro is different as it does not want to replay old text, it wants to remember the meaning of what happened.

4hViews 24Likes 0Bookmarks 0
Rohan Paul@rohanpaul_ai

The nightly memory pipeline starts by reading the day’s signals: chat history, file-system events, and app activity.

Then it extracts typed entities like Projects, People, Decisions, Preferences, and Open Tasks, which are the pieces that usually matter again tomorrow.

Rohan Paul@rohanpaul_ai

Kocoro works like Mac-native AI coworker: you type normal requests, and it can read files, edit code, schedule reminders, connect to tools like GitHub or Slack through MCP, and run background agents for recurring work.

4hViews 21Likes 0Bookmarks 0
Rohan Paul@rohanpaul_ai

The super useful part is the graph structure.

A graph means Kocoro does not only store “task X exists,” but can connect it to a project, a person, a decision, and a next action, so the agent can recover the working context without asking the user to paste everything again.

Rohan Paul@rohanpaul_ai

The nightly memory pipeline starts by reading the day’s signals: chat history, file-system events, and app activity.

Then it extracts typed entities like Projects, People, Decisions, Preferences, and Open Tasks, which are the pieces that usually matter again tomorrow.

4hViews 19Likes 0Bookmarks 0
Aasim Mahmood | ₿@K9Aasim

Safety first, trillion-dollar valuation second. 🤔

Anthropic just dropped Claude Fable 5, the "Mythos-class" model that's so powerful it comes with heavy guardrails on cybersecurity, biology, and chemistry. It's like buying a Ferrari with a 30mph speed limiter. 🏎️💨

Why the caution? The unshackled Mythos version proved it could find thousands of critical cyber vulnerabilities on its own. Now, with a 10.9B in Q2 revenue, Anthropic is balancing breathtaking ambition with genuine safety concerns ahead of its upcoming IPO.

#Anthropic #ClaudeFable #ResponsibleAI #Valuation

4hViews 20
AIKeaton UK@ByteFuserUK

@rohanpaul_ai Yep, this is the real problem for me. Not raw model IQ, just remembering the right stuff without turning into a junk drawer.

4h