{ "version": "https://jsonfeed.org/version/1.1", "title": "GainWix AI Studio blog", "home_page_url": "https://gainwix.ai/blog/", "feed_url": "https://gainwix.ai/feed.json", "description": "Notes on governed AI — how we think about control, accountability, and keeping AI in real work.", "language": "en", "items": [ { "id": "https://gainwix.ai/blog/gainwix-0-1-16-second-brain/", "url": "https://gainwix.ai/blog/gainwix-0-1-16-second-brain/", "title": "GainWix 0.1.16: an AI that already knows your work", "summary": "Give your AI memory that lives in files you own. Release 0.1.16 fills your Second Brain by asking you one question at a time.", "content_text": "Your AI can hold real memory of your work, but only if that memory lives in files you own instead of inside a chat window. GainWix gives you an AI avatar with a Second Brain: linked notes about you, your projects, and the people around you, kept as plain files on your own machine and drawn as a map you can see. Release 0.1.16 is about filling that map without it turning into homework. \n Every morning, a stranger \n Open a fresh chat and the work starts before the work. Who you are. What your company sells. Which project is live this week, which one you parked, and what you decided last Thursday and why you decided it that way. You type all of it out. Then you type it again tomorrow, in slightly different words, to an assistant that has no idea you have met before. \n The tools do have an answer for this. Create a project. Write a system prompt. Tell it to remember. Every one of those is homework, and almost nobody keeps homework current past the second week. The project description still names a client who left in March. The system prompt describes a job you finished before that. So you stop trusting it and go back to re-explaining yourself, which is the thing you wanted to stop doing. \n A colleague doesn't work like that. They know you. Nobody briefs a colleague every morning on what the company does, because last quarter is still in their head and so is the argument you both had on Tuesday about the launch date. Ask them a question and the answer already has your context in it. The difference between that and a fresh chat isn't intelligence. It's whether the memory belongs to you or to a window you closed last night. \n A Second Brain you can see \n Your avatar's memory is a Second Brain. Notes about you, your work, the people you work with, and the projects in front of you, each one linked to the notes it relates to. One note per thing worth remembering. Your company gets a note, and so does every client, every product you sell, every person you deal with often, and every decision you'd hate to have to work out again from scratch. \n That last kind is the one people forget to keep. Most of what makes you quick at your own job isn't facts, it's settled arguments. Why you set the rate you did. Which supplier you stopped using and what went wrong that week. A note holding the reason, linked to the person and the project it came out of, is worth more to your avatar than a folder of tidy documentation nobody has opened since it was written. \n Notes are files, not rows in a database \n Open the Second Brain folder and there's nothing clever in there. Files. One per note, in Markdown, named so you can tell what they are without opening them. A few lines at the top say what kind of note it is and when it last changed. The body is a handful of sentences in ordinary English, and the links to other notes are written into the text itself, so the connection between the client note and the project note is something you can read rather than something you have to trust. \n That plainness is the reason for storing them this way. Open a note in any editor and fix a sentence. Search the folder from a terminal. Copy it to a backup drive, keep it in a git repository, or read it in five years on a laptop that has never had GainWix installed. Nothing is locked inside a store you can't open, and nothing has to leave your machine for you to look at it. \n Why the map matters \n The app draws every note and every link as one picture. Notes are dots, links are lines, and the shape of your work appears as the collection grows. It looks good on a screen, and that isn't why it's there. \n The map is how you see what your avatar believes. A dot sitting on its own means you mentioned something once and it never got connected to anything. Two dense clumps with nothing running between them usually means the avatar hasn't worked out that your biggest client and your newest project are the same story. Both of those are visible in a glance, and both are fixed by saying so out loud. \n Memory you can't see is memory you can only test. You find out what it forgot when it gets something wrong in front of a customer. A picture you can read turns that into something you check on a Sunday evening and correct in one sentence. \n Chat and Talk \n The avatar's home is two places now, named for what they are. Chat and Talk. \n Chat is where you type. Talk is where speaking lives. Both open onto the same avatar and the same Second Brain, so a note made in one is already known in the other. There's no second copy of you to keep in sync, and no setting to decide which half of your context each place gets. \n Ask your avatar to interview you \n Empty notes are why most second brains die in week one. Staring at a blank file and being asked to describe your own working life is a bad first task, so this release hands that part to the avatar. \n Ask it to interview you. It asks one question at a time. \n It asks something small and concrete. What the company does, who your customers are, what you're shipping this quarter. You answer in a sentence or two, the way you'd answer a colleague at a desk. It writes a note, links it to the notes it fits with, and the new dot appears on your map while you watch. It asks the next question, shaped by what you already said. You stop whenever you want. Nothing is half-finished, because every answer was already saved as its own note. \n Ten minutes on a Friday afternoon is enough to change what the following Monday feels like. There's no form, no onboarding wizard, no fifty-field profile to complete before the thing becomes useful. You shouldn't have to work at being known. \n Come back a week later and the second session is a different conversation. The avatar has read what you already told it, so it stops asking about the shape of your business and starts asking about the edges. Who signs off on an invoice. What you do when a client goes quiet for a month. The questions get narrower because the map under them is denser, and the notes that come out of them are worth more for the same reason. \n By the third time you're not really building anything. You ask your avatar an ordinary question about your own work, and the answer arrives with your context already in it, because it read your notes before it said a word. That's the point where the folder stops feeling like a chore. \n It reads your notes before it answers \n Every request your avatar handles starts the same way. Before it answers, before it touches a single tool, it reads what it knows about you. Ask it to draft a reply to a client and it already has who that client is, what you sold them, and what you agreed in June. You attach nothing. You paste no brief. \n The same thing happens on the way back out. When something worth keeping comes out of the work, a decision, a name, a figure you had to go and look up, it becomes a new note linked into what was already there. The memory gets better because you used it. A system prompt does the opposite, quietly going stale from the day you wrote it. \n What's new in 0.1.16 \n Three new things, three that got better, three that were broken. \n Change Kind What it means Chat and Talk New The avatar's home is two places, named for what they are. Chat is where you type, Talk is where speaking lives Interview me New Ask from either place and your avatar takes you one question at a time, writing each answer into your Second Brain as a linked note What's new, in the app New The release notes live inside GainWix, so you can read what changed in the version you're actually running A bigger, calmer window Improved More room to work, and GainWix branding on every screen Guided tours Improved A screen can explain what it's for, and a ? in the corner replays the tour whenever you want it again One settings screen Improved Settings used to be split across two places. Now there's one, with the recycle bin beside it Sign-up codes Fixed Codes arrive reliably Editing a note Fixed Updating a note keeps its connections on the map instead of dropping them The gateway card Fixed It no longer turns up on screens where it doesn't belong \n It runs next to Claude, on your machine \n GainWix AI Studio is a local-first control plane for Claude Desktop, Cowork, and Claude Code. Every Claude tool call goes through one gateway you control, with per-tool allow-lists and budgets, an Approval Inbox for risky actions, and a tamper-evident audit log. Connecting your Claude apps is one step, and after that the Second Brain your avatar builds is the same memory those apps read from. \n The plumbing under that is the Model Context Protocol , the open standard Anthropic published for connecting AI apps to tools and data. One protocol, many servers, one door in front of all of them. \n If the gateway part is new to you, What is an MCP gateway explains the door itself, and One doorway is the story of a first real handoff. The GainWix Workmates page lists the roles you can hire. \n Get 0.1.16 \n The macOS build is on the download page . The app is free. So are the Claude courses inside it. \n Install it, open your avatar, and ask it to interview you before you do anything else. Answer five questions while your tea goes cold. Tomorrow morning, open Chat and ask it something about your own business, and watch it answer without asking you who you are.", "image": "https://gainwix.ai/assets/blog/gainwix-0-1-16-second-brain.jpg", "date_published": "2026-09-04T00:00:00Z", "date_modified": "2026-09-04T00:00:00Z", "authors": [ { "name": "The GainWix team" } ], "tags": [ "second-brain", "memory", "notes", "release", "avatar" ] }, { "id": "https://gainwix.ai/blog/what-is-an-mcp-gateway/", "url": "https://gainwix.ai/blog/what-is-an-mcp-gateway/", "title": "What is an MCP gateway? One door for every Claude tool call", "summary": "An MCP gateway is one controlled doorway that every AI tool call passes through, so you can log it, block it, or ask a human before it runs.", "content_text": "An MCP gateway is one controlled doorway between an AI app and every tool it can reach. Claude Desktop, Cowork, or Claude Code sends a tool call. The gateway sees it first. It writes the call down, lets it pass, stops it, or holds it for a human. Nothing touches your files, your database, or your email without going through that door. You own the door, so you own what the assistant can do. \n What MCP is, in plain words \n MCP stands for Model Context Protocol. Anthropic published it as an open standard in November 2024. The announcement describes the problem it fixes. Every new data source used to need its own custom connection. MCP replaces those one-off connections with a single protocol. \n The MCP site uses a plain picture. MCP is like a USB-C port for AI apps. One plug shape, many devices. \n Three parts make it work. The host is the AI app, such as Claude Code. Servers are the small programs that offer tools, files, and prompts. Clients are the connectors inside the host that talk to each server. \n The word that matters here is \"tools\". In MCP, a tool is a function the model can run. Read a file. Query a database. Send a message. Delete a record. Each one is real work in your real world. \n Why every tool is a door \n The MCP specification says this outright. Tools \"represent arbitrary code execution and must be treated with appropriate caution\". It goes further. Hosts \"must obtain explicit user consent before invoking any tool\". \n The Claude Code MCP docs say the same thing from the other side. \"Verify you trust each server before connecting it.\" Servers that fetch outside content can carry prompt injection. A web page can hide an instruction. The model reads it. The model might act on it. \n So each MCP server you connect is a door into your house. Some doors lead to the kitchen. Some lead to the safe. The model doesn't always know the difference. You do. \n Why one doorway beats many \n Picture a normal setup. Claude Code has five MCP servers connected. GitHub, a database, Slack, the file system, and a browser. Each server is its own door. Each one has its own trust dialog, its own permissions, its own idea of what's risky. \n Now something goes wrong. A tool call you didn't expect. Which door did it come through? What did it touch? Was it the browser reading a page that told the model to do it? You'd have to check five places. Five logs, if they exist at all. \n An MCP gateway collapses those five doors into one. Every call from every server passes through the same point. One set of rules. One log. One inbox for the calls that need a human. \n The spec admits MCP \"cannot enforce these security principles at the protocol level\". The protocol carries the messages. Somebody has to stand at the door. That somebody is the gateway. \n The story in One doorway between Claude and everything you own shows what this feels like on a working Tuesday. This post is the plain version of the same idea. \n What an MCP gateway logs \n The gateway sees everything, so it can write everything down. A good log records the same fields for every call, no matter which server handled it. \n The time the call arrived. The tool that was called, and the server it belongs to. The arguments the model passed in. The file path, the query, the message text. What the gateway decided. Allowed, blocked, or sent for approval. Who approved it, if a person did. What came back. Success, error, or refusal. \n The log has to be tamper-evident. That's a plain phrase for a simple promise. If any line is edited or removed later, you can tell. Without that, the log is a story. With it, the log is a record. \n What an MCP gateway blocks \n Blocking works on two levels. The first is the allow-list. You name the tools the model may use. Everything else is off. A new server might offer twenty tools. You allow four. The other sixteen don't exist as far as the model is concerned. \n The second level is the budget. Some tools are fine in small doses and worrying in large ones. Running the test suite once is normal. Running it two hundred times in a loop is a bug or a runaway. A per-tool budget puts a ceiling on it. Hit the ceiling, and the gateway stops the call. \n Blocking is quiet. The model gets a refusal and moves on. You get a line in the log. Nobody has to wake up. \n What an MCP gateway asks a human about \n Some actions shouldn't be blocked and shouldn't be automatic either. Sending an email to a client. Pushing to the main branch. Deleting a table. Posting in the team channel. These are the calls where a human should look first. \n A gateway holds these calls in an inbox. The model pauses. You see what it wants to do, in full. You approve, or you don't. Then the model continues or takes another path. \n This is the \"explicit user consent\" the spec asks for, made practical. Not a dialog on every tool call. A dialog on the ones you marked as risky. \n How GainWix AI Studio does it \n GainWix AI Studio is a local-first control plane for Claude Desktop, Cowork, and Claude Code. Every Claude tool call goes through one gateway you control. It runs on your own machine, not on somebody else's server. \n Inside it, you get the three things this post has described. \n Per-tool allow-lists and budgets. You pick which tools Claude can use and how much of each. An Approval Inbox. Risky actions stop there and wait for you. A tamper-evident audit log. Every call, every decision, written down and hard to quietly change. \n The app is a free desktop download from gainwix.ai for macOS, Windows, and Linux. The security page lists what the gateway does and how it stores the log. \n GainWix Workmates run through the same door. A role you hire, not a feature you configure, but still one doorway. \n A worked example with Claude Code \n Say you ask Claude Code to fix a failing test, push the fix, and tell the team. Here's what a day with the gateway looks like, call by call. \n Tool call Gateway action What you see Read files in the project Allow One line in the audit log Run the test suite Allow, inside its budget A log line and a count against the budget Push to a feature branch Allow A log line with the branch name Push to the main branch Ask An item in the Approval Inbox Drop a database table Block A blocked entry in the log, and Claude gets a refusal Send a message to the team channel Ask The full message text waiting in the Approval Inbox \n Claude Code does the real work. It reads, tests, and pushes without asking. It stops at the two doors you marked. You glance at the inbox between meetings, read the message, and tap approve. Later, anyone can read the log and see every step in order. \n That's the whole idea. The assistant isn't slower at the safe parts. It's only slower at the parts that should be slow. \n MCP gateway or MCP server \n People mix these up, so here's the short version. \n An MCP server offers tools. It's the thing on the far side of the door. GitHub's server offers GitHub tools. A database server offers query tools. \n An MCP gateway sits on the near side, between the AI app and all of its servers. It doesn't offer tools of its own. It decides what happens to each call. You can have many servers. You want one gateway. \n Where to start \n Download the GainWix app and connect Claude Code. Open the allow-list and tick the tools you're comfortable with. Mark send, delete, and push-to-main as ask. Then hand Claude Code a real job and go make tea. \n When you come back, the Approval Inbox has one item in it. The message to your team, word for word, waiting for your nod. Everything else already got done, and the log says so. If you want the longer story of that first handoff, read One doorway next.", "image": "https://gainwix.ai/assets/blog/what-is-an-mcp-gateway.jpg", "date_published": "2026-09-03T00:00:00Z", "date_modified": "2026-09-03T00:00:00Z", "authors": [ { "name": "The GainWix team" } ], "tags": [ "mcp", "gateway", "claude-code", "approvals", "audit-log" ] }, { "id": "https://gainwix.ai/blog/one-doorway/", "url": "https://gainwix.ai/blog/one-doorway/", "title": "One doorway between Claude and everything you own", "summary": "A short story about handing Claude Code a real job — and staying in control of every door it opens.", "content_text": "On a Tuesday morning, Meera asks Claude Code to do something real. Not a toy prompt — an actual job. “Clean up last quarter's customer records, then draft the summary the finance team keeps asking for.” It is exactly the kind of work she would love to hand off. It is also exactly the kind of work that used to keep her from handing anything off at all. \n Because to do that job, Claude Code has to reach into her world. It needs her files. It needs the customer database. It needs the tool that sends the finance team their email. The moment an assistant can touch those things, a quiet question sits behind every request: what else can it reach, and would I even know? \n The old way: a house with every door open \n Before GainWix, connecting an assistant to real tools felt like leaving for work with every door in the house unlocked. Maybe nothing goes wrong. Probably nothing goes wrong. But you spend the whole day not quite sure, and so you never really leave. Meera did the same thing everyone does: she kept the interesting work to herself and gave the assistant only the safe, boring bits. \n What GainWix AI Studio changes \n GainWix AI Studio puts one doorway between Claude Code and everything Meera owns. Instead of the assistant wandering the house on its own, every reach for a file, the database, or the email tool goes through that single door — and Meera holds the keys to it. \n She decides, once, what lies beyond each door. The customer database? Claude Code may read it, never change it. Her files? It may look, but writing anything needs a nod from her. Sending email? That door stays shut until she opens it herself. She sets this up in plain terms in a few minutes, and then she can forget about it, because the studio remembers for her. \n The assistant still does the work. It just does it through a door you own, on the terms you set. \n Back to Meera's Tuesday \n Claude Code gets to work. It reads the messy records — that door was open, so it simply walks through, and the studio quietly notes that it did. It tidies them up in a draft, not in the real database, because the door to changing customer data was one Meera left closed. When it is ready to send the finance team their summary, it stops at the last door and waits, holding the finished email up for her to see. \n Meera glances at it between meetings, reads the summary, and taps approve. The email goes. The whole time, she was not watching over the assistant's shoulder — she was doing her own work, trusting that anything worth her attention would come to her, and everything else would simply, safely, get done. \n And if she ever wonders what happened \n Later, someone asks what the assistant actually touched that morning. Meera does not have to guess. The studio kept a plain, unbroken record of every door Claude Code opened and everything it did — the reads it made on its own, and the one send she approved. She can read it back at any time. The control was never a feeling. It was written down. \n The point of the story \n Claude Code is capable enough to do the real work. What was missing was never the intelligence — it was the doorway. Give the assistant one entrance you control, decide what lies behind each door, and the fear that kept the interesting work to yourself simply goes away. You stop guarding the house, and you start getting the day back.", "image": "https://gainwix.ai/assets/blog/one-doorway.jpg", "date_published": "2026-07-14T00:00:00Z", "date_modified": "2026-07-14T00:00:00Z", "authors": [ { "name": "The GainWix team" } ], "tags": [ "claude-code", "gateway", "approvals", "audit-log" ] } ] }