Insights from your AI chats
Runwork Chats puts your local AI conversations in one place, and your own AI reflects on each one, turning what it finds into skills and automations worth keeping.
You walked through the same refund-handling steps in three conversations this week. Save it once as a team skill, and everyone's AI knows it.
Analyzed on this machine. Content never leaves it. What Chats & Insights gives you
Every conversation, one list
The Chats view shows all your local AI conversations across Claude Code, Codex, Cursor, Claude Desktop, and Gemini CLI: agent, project, message and tool-call counts, and when you last touched it. Saved checkpoints sit in the same list, so your whole AI life is in one place.
Resume, share, or analyze from one pane
Open any conversation to see its session stats and act on it: resume it in the tool it came from, hand it to a teammate through Share & Resume, or send it straight to analysis without waiting for the queue.
Reflection minutes after the conversation ends
When a conversation ends, Runwork notices and queues it for analysis on your machine. The suggestion arrives while you still remember the work, not in a weekly report about things you did last Tuesday.
Suggestions drawn from real work
The Suggestions tab collects what reflection found: a walkthrough you keep repeating that should be a team skill, a manual chore that could be an automation, a capability you have not touched yet. Each one comes with its evidence and a concrete next step.
Analysis on your machine, by your own AI
Reflection runs through the AI tool you already use, on your device. Conversation content is read locally and stays local. Runwork's servers only ever see metadata: a capability label, a timestamp, a confidence score. A test suite pins that invariant.
A triage pass that guards your subscription
Every finished conversation gets a quick, cheap triage first. Only the ones with something worth finding get the deep look, under a daily budget you control, with an "Analyze now" override when you want the backlog drained. Every run records which model it used and what it consumed.
Suggestions where you will actually see them
A tray line when new suggestions land, a notification when a run finishes, a banner on home, rows in the team activity feed, and an optional email digest. On the web, suggestions appear as metadata pointers that open the full content in the desktop app.
macOS, Windows, and Linux
Conversation detection, analysis, and suggestions work the same on all three platforms, terminal-only included: the CLI owns the whole engine, and the desktop app is one way to drive it.
Access from anywhere
Desktop & Web
Codex DEFAULT Claude Code Cursor Open Chats in the Runwork desktop app for the full experience. On the web, the Conversations page shows your suggestions as metadata pointers that open in the desktop app.
CLI
Why Chats & Insights matters
- One view across five different AI tools; nothing else has this
- Analysis runs on your machine with your own AI, never on our servers
- Suggestions arrive minutes after the conversation, while context is fresh
- Cheap triage first, deep analysis only when warranted, budget in your hands
- Every suggestion is one step away from a saved team capability
- Conversation content never leaves your device, and a test pins it
How Chats & Insights works
It starts with the conversations you already have
Your AI tools keep transcripts on your machine. Runwork reads that record locally and builds one list: every conversation, from every tool, with its project, message counts, and tool activity. Checkpoints you saved and conversations teammates shared live in the same place, so "where was that thread?" stops being a question.
Reflection happens right after the work
When a conversation goes idle, it counts as finished and joins a local queue. First comes triage: a fast, cheap pass by your own AI that answers one question, is there something here worth a closer look? Most conversations end there, at near-zero cost. The few that matter get the deep analysis, which reads the full transcript locally and writes up what it found.
What it finds is a suggestion, not a report
Reflection does not summarize your week. It points at something specific: this walkthrough you have now done three times should be a team skill; this manual step could be a schedule; this capability exists and you have never used it. Each suggestion carries its evidence and a next step, and acting on it puts a real capability into your team's workspace.
Privacy is structural, not a policy
The analysis runs on your device, through the AI subscription you already have. Insight content is stored on your machine and rendered in the desktop app. What syncs to Runwork is metadata only: a capability label, a timestamp, a source project name, a confidence score. The web app renders those as locked pointers that deep-link into your desktop. There is no server-side copy of your conversations to protect, because one never exists.
And you can see what it cost
Every reflection run records which tool and model served it and what it consumed, so there is no mystery about what analysis is doing with your subscription. A daily budget caps deep runs, and "Analyze now" lets you drain the backlog on demand.
Frequently Asked Questions
Does my conversation content reach Runwork's servers?
What does reflection cost to run?
Which AI tools are covered?
Can I turn reflection off?
Can my manager read my conversations?
How is this different from Share & Resume?
Use Cases
Related Features
See How Teams Use Chats & Insights
Ready to try Chats & Insights?
One shared cloud under the AI tools your team already uses.