Sprint planning
Meeting · 12 Sept · Ana Ionescu
Guest checkout goes to phase 2. The date doesn't move. I'd rather ship a smaller checkout on time than a bigger one late.
Meetings, Slack, GitHub and Jira flow into one memory for each project. Everyone on the project asks the same chat, and every answer carries the quote and the source behind it.
A copy of the product answering a question, shown as text: the question, the answer, the three sources it cites, and one of those sources opened to the sentence the answer came from.
From the demo project this page is built on.
01 Learns from
Connect a source once. What was said there becomes memory anyone on the project can ask about, each entry holding the sentence it came from.
02 Who asks
The people on a project need different things from it. Each of them asks in plain words and gets an answer built only from what they are allowed to see.
“Are we still on track for the 15 October launch?”
Answer cites:Leadership sync · 3 Sept
“What did we promise the client about guest checkout?”
Answer cites:Client call · 9 Sept
“Why was sign-up cut down to a single step?”
Answer cites:#design · Slack
“Which payment tests have to pass before code freeze?”
Answer cites:PR #212 · GitHub
“What changed in checkout since the last test run?”
Answer cites:QA-41 · Jira
“When will guest checkout be live?”
Answer cites:Sprint planning · 12 Sept
03 How it answers
Studio is building a checkout for its client Acme. Guest checkout moved to phase 2, not everyone agreed, and the client is in the same chat as the team.
Why isn't guest checkout in the launch?
It moved to phase 2 at sprint planning on 12 September. Building it for launch is about six more developer days, and it takes the payment tests from 24 to 48 cases before code freeze.
Sara disagreed on the day, on conversion grounds. The 15 October date cannot move, because Acme's campaign starts on 20 October.
Elena's answer is not the model being careful. Her search cannot reach those rows at all: who sees what is a Postgres policy, checked before anything is sent to the model.
04 How it works
What goes in, what memory keeps, how you ask, and what is still owed.
05 Under the hood
The parts nobody sees in a demo, and the first things a company asks about.
Postgres row-level security checks every query, so a client never reaches a team-only row, not even by mistake.
It runs on your server or ours in Europe, and your content goes only to Anthropic, on your own key.
The whole app is in both languages, and anyone can ask in the one they think in.
Each company chooses how long things are kept. A nightly job removes the rest, without anyone remembering to.
Keys, people, roles, sources, sharing and retention changes are recorded with who made them and when.
Anthropic keys and connector tokens are encrypted at rest, never logged and never sent to the browser.
06 Where to use it
Open InteliLang in the browser, or ask the same project memory from Claude, Claude Code, Cursor or VS Code without switching windows.
The whole app: the chat, memory, the board and files, for everyone on the project, including your client.
Who pays for the AI: Your company's Anthropic key, at Anthropic's price
Add InteliLang as a connector in Claude on the web, desktop or phone, then ask about a project in the middle of any conversation.
Who pays for the AI: Your own Claude plan
Connect it with one command and a personal key. Check what was decided before you write the code for it.
Who pays for the AI: The account you use Claude Code with
Paste a few lines into the editor's MCP settings, and its assistant can read the project's decisions next to your code.
Who pays for the AI: The editor's own AI
Every way in sees only what you can see on that project. A personal key can be revoked in one click, and the app it connected stops right away.
07 Numbers
No customer counts and no made-up benchmarks. Just how the product is built to spend your money.
08 Who it's for
Studios, agencies and product teams that work in projects, often with the client in the same conversation. It is not trying to be a search engine for a company of ten thousand.
Create a company, paste your Anthropic key, connect a Slack channel or upload a meeting. No sales call and no rollout project.
The AI runs on your own key, billed by Anthropic with no markup. Questions the project has already answered cost nothing.
Invite the client as a guest. They ask the same chat and get true answers, built only from what they are allowed to see.
Runs on your server or ours, in the EU. Old content is deleted on the schedule you choose.
Connect a Slack channel, a repository or one recorded meeting, then ask it something you already know the answer to. The answer should come back with the quote you were thinking of, and the source next to it.
May we count visits to improve InteliLang? We would keep a small cookie for it. Nothing you type is ever sent. Privacy