Conduct gives every team lead spend visibility, policy enforcement, and a full audit trail — across every AI coding tool your team uses. Plus 22 ready-made agent workflows, DORA-lite metrics, and per-agent scorecards — so you can see what's working, not just what's running.
Connect your stack, install a playbook, and your first agent run goes through Guard automatically.
Link GitHub and Slack. Add credentials in Settings → Environments. No migration, no new tooling — Conduct wraps around what you already use.
Browse 22 pre-built agent playbooks. Click Install — the canvas is pre-wired. Add your credentials and the agent is ready to run.
PRs, reviews, triage comments, incident diagnoses — delivered to Slack. Every run passes through Guard automatically. Approve or reject in one click.
Stop doing the same work around every PR. 22 pre-built playbooks handle issue triage, code review, incident response, and release notes — with human approval before anything ships. Every agent gets an A–F scorecard and DORA-lite metrics so you know what's actually delivering value.
Browse Marketplace →Your developers are spending money on AI tools you can't see into. Guard gives you real-time spend per developer, policy enforcement, and a full audit trail — across every AI tool, from one dashboard.
See the full Guard story →Agents that improve with every run. Before acting, each agent recalls what it learned on this repo. After the run, it records the outcome. The 10th PR is smarter than the first.
Memory docs →Conduct sits between your existing tools and your AI agents — routing events through agent templates, enforcing Guard policies, and recording every decision.
Agents and external tools connect via the Conduct MCP server — query runs, trigger workflows, and check Guard policy state from any MCP-compatible AI tool. conduct mcp install wires it in 30 seconds.
18 pre-built agent templates across 10 categories. Browse, install, configure credentials, run.
Issue labeled ai-ready → AI implements fix, runs tests, opens PR → human approves in Slack before merge.
PR opened → AI reviews the diff for bugs, security issues, and style → posts a structured review comment.
PR opened → OWASP Top 10, secrets, auth bypass scan → posts a security findings report to the PR.
Build fails → AI diagnoses the failing step, finds the root cause → posts a structured fix to Slack.
New issue opened → AI classifies type and priority, adds labels, asks clarifying questions if vague.
Alert fires → AI correlates recent deploys and commits → posts a root-cause hypothesis to #incidents.
AI spend is now a board-level concern. Guard gives your VP Eng and CFO spend visibility, policy enforcement, and a full audit trail — across every AI coding tool your team uses.
Which developer spent $800 on Claude last month? Which project burned your budget? Today, you can't answer these questions.
Developers run destructive commands in production environments, push secrets to AI context, deploy without review. Uncontrolled.
When something goes wrong, there's no record of what the AI did, what was approved, or who authorised it.
Define spend caps, blocked commands, approved tools, and enforcement mode in the Guard dashboard. One config. Applied automatically to every developer.
Team lead invites developers from the dashboard. Developers log in — policies download and apply to their local environment automatically. No IT ticket. 30 seconds.
conduct login
The conductguard-mcp server connects Guard to Claude Code, Cursor, Windsurf automatically.
Most AI tools start from scratch every time. Conduct agents remember. Before each run the agent recalls what it learned on this repo. After the run it records the outcome.
Before the agent block runs, a Memory read block retrieves the most relevant past summaries using vector similarity — not just recent history, but what's most applicable to this exact task.
After the run, a Memory write block stores what was done: the approach, files changed, what worked. That becomes context for future runs on the same repo.
Each playbook builds its own knowledge store per repo. What Autopilot learns on conductai/api never bleeds into conductai/web. Two scopes: repo or workspace.
Without a spec, AI coding tools drift. Requirements get lost. Nobody can trace which code maps to which decision. SDD gives every AI action a why — and Conduct enforces it automatically.
Describe what you're building. The agent asks 3 questions and writes a structured SPEC.md with numbered, testable FR-xxx requirements. No account needed.
Generate SPEC.md →sdd-bootstrap reads your SPEC.md and commits 6 files to your repo — AGENTS.md, DESIGN.md, PLAN.md, SPRINT.md, CLAUDE.md, and the spec index. Fully wired in minutes.
See the full workflow →A pre-merge hook blocks any PR where changed files have no FR reference. The only platform that enforces spec traceability at the git layer — not just stores requirements.
Learn about enforcement →Open source tools for developers. An honest comparison with every major AI engineering tool.
Agent Booster cuts per-request AI costs by up to 15× by sending only the relevant code. Claude Code Team Kit gives your whole team a shared setup in one command. Both free, MIT licensed.
Browse open source tools →How does Conduct compare to GitHub Copilot, Devin, CodeRabbit, LinearB, and others? Honest feature matrix, strengths, trade-offs, and a decision guide for your team's situation.
Compare all tools →"Labeled a bug at 5pm on a Friday. PR was open by 5:02pm, tests green, ready to review Monday morning. That's the thing — it just works while we're not watching."
"Guard paid for itself in week one. I could finally see exactly which developers and which tools were burning our Claude budget. We cut spend by 40% without slowing anyone down."
"The Copilot reviewer playbook caught three hallucinated API calls in AI-authored PRs that our human reviewers missed. We won't merge AI code without it now."
Sign in, connect a GitHub repo, install your first playbook. Running in under 5 minutes.