Onboarding
Before: 2–3 weeks to find your way around, then 2–3 more at half speed
New developers pick up tasks on day one
Made on
Tilda
AI Council gives Claude Code, Codex and other MCP agents what your senior engineers know: the code graph, the decisions behind it and the rules your team follows. The agent gets it before it edits a file, and you can see what it did with it.
From $20 per user per month. No sales calls. Keep your IDE, agent and tracker. Runs in our cloud or on your own servers, including air-gapped networks.
A 12-minute walkthrough on Keycloak, an open-source project you can check yourself. Chapters
It sits between your code and your AI agents. It keeps a knowledge base about the project, fills it while the team works and checks that it still matches the code. Nobody has to go and update a wiki.
A code graph, team decisions with the reasons behind them, and MUST and SHOULD rules, all linked to modules and functions. Knowledge is scoped to git branches, rules apply to the whole project. Agents get what they need through hooks and MCP without asking.
Tickets from GitHub Issues and Redmine are analyzed automatically: root cause, affected files, status. You can connect project documents and related repositories. Answers from the project chat are saved as knowledge, and agents and people record decisions as they work.
Code is re-analyzed after changes. When the code behind an entry changes, a similar entry appears or a branch is merged, the entry goes to review: a model labels it with a quote from the source, and a person can override the label. Rules are checked in the diff and in MR review.
Across 700+ companies that adopted AI agents, code volume grew 30% while the number of closed tasks stayed flat. The time went into review and rework. Harvard, 2026
Before: 2–3 weeks to find your way around, then 2–3 more at half speed
New developers pick up tasks on day one
Before: the agent doesn't know team conventions and does things its own way
Team rules reach the agent before it edits, and the diff is checked for violations
Before: agent code sits in review because nobody trusts it
The MR is checked against the same rules, and the reviewer sees what was broken and where
Before: decisions live in people's heads and leave with them
Decisions and their reasons live in a shared knowledge base linked to the code
AI Council sent the agent this before it edited billing/invoice_sender.py
payments-retry queuelegacy/gateway_v1AI Council builds a code graph and links team decisions to modules and functions.
Through hooks and MCP, the relevant decisions and rules arrive when the agent opens or changes a file.
Decisions made during a task go into the shared knowledge base, kept on their git branch until merge.
instead of 2–3 weeks before a new developer starts closing tasks. Compared with the last few hires before AI Council
less time and fewer tokens on a single task. Clean test: same task, same project, same agent, with and without AI Council
less time and fewer tokens across a series of tasks when several developers and agents work in parallel
Startup is $30 per user per month for up to 10 users, Standard is $20 per user for 10–50. Cancel any time.
GitLab or GitHub. AI Council analyzes the code and your tickets and builds the project graph.
One command installs the hooks and the MCP server for Claude Code or Codex on each developer's machine.
Need more than 50 seats, SSO or an air-gapped install? Message us on Telegram or LinkedIn.
If you maintain an open-source project or teach programming at a school or university, AI Council is free. Apply here