01
Fragmented provider CLIs
Every provider has different mutation rules, different APIs, and different edge cases.
TRUST LAYER FOR AI AGENTS
SAFE PRODUCTIONCHANGES FORAI AGENTS
Keycli sits between agent intent and production mutation: turning risky infra changes into structured plans, policy checks, approvals, provider-backed execution, and audit trails.
LIVE WHEN CONNECTEDAPPROVAL AWAREAUDIT FIRST
Works across provider surfaces like Vercel, GitHub Actions, and Render — without giving agents prod god mode.
CONTROL PLANE :: LIVE / GOVERNED
intentrotate STRIPE_SECRET across vercel + github-actions and deploy
riskhigh / approval_required
executionprovider-api when connected / simulated when not
providersvercel :: github-actions :: render
INTENT→PLAN→POLICY→APPROVAL→APPLY→AUDIT
WHY NOW
Agents are getting powerful faster than production workflows are getting safe
Today, agents can write code, open PRs, and drift toward deploys, config updates, and secret mutation. The operational workflow behind those actions is still fragmented.
02
Scattered approvals
Humans approve changes in whatever system happens to be nearby, with weak shared context.
03
Brittle secret rotation
Cross-provider key changes are easy to get half-right and hard to prove after the fact.
04
Disconnected deploy workflows
Config changes and deploy triggers rarely feel like one governed operation.
05
Thin auditability
Teams can see that something happened, but not always who approved what and why.
06
Agent over-permissioning
Without a trust layer, agents are either underpowered or handed broad production authority.
CONTROL PLANE FLOW
One control plane between agents and production
Instead of letting agents mutate infrastructure directly, Keycli gives them a constrained path with explicit state at every step.
REQUEST
Intent
The agent expresses the change it wants, not arbitrary shell access.
STRUCTURED
Plan
Keycli turns intent into a structured change plan with real targets and diff context.
CHECKED
Policy
Risk and readiness are evaluated before the system acts like a mutation is safe.
REQUIRED
Approval
Humans approve when the plan crosses the threshold for governed live change.
SCOPED
Apply
Execution runs through provider adapters, not direct agent access to production.
RECORDED
Audit
The resulting run, approvals, and outcomes land in one audit trail.
FIRST WEDGE
Safe config and secret changes across real providers
The first useful wedge is not “AI does everything.” It is governed config and secret mutation across real provider surfaces.
Rotate a shared secret across Vercel + GitHub Actions
Plan one change, inspect provider targets, require approval when needed, then apply through connected adapters.
MULTI-PROVIDERHIGH SIGNALAUDITED
Update preview env vars safely
Handle low-risk preview changes through a consistent flow instead of ad-hoc provider commands.
LOW RISKVERCEL READYFAST LOOP
Approve in GitHub, inspect the run in Keycli
Use GitHub comment approval capture for the current narrow wedge, then inspect the resulting plan and audit trail.
GITHUB FLOWCURRENT WEDGETRUTHFUL
POSITIONING
Not a vault. Not a wrapper. A trust layer.
Keycli is better understood as agent-safe production change orchestration than as one more dashboard around secrets.
NO / WRONG FRAME
Not another secret manager
The product is about governing live changes, approvals, and execution — not selling static storage alone.
NO / WRONG FRAME
Not a wrapper around provider CLIs
The value is consistent plan, policy, approval, apply, and audit semantics across providers.
YES / TRUST LAYER
Built for agent-native workflows
Plans, next actions, approval gates, scoped execution, and auditability are designed around AI-assisted operations.
TECHNICAL PROOF
Already grounded in a real wedge
This landing stays tight to the product truth today: a hosted control plane with real provider execution when the right connections exist.
- +Hosted control plane with plans, approvals, runs, and audit.
- +Live provider execution for Vercel, GitHub Actions, and Render when properly connected.
- +Mixed-provider execution on supported targets.
- +GitHub-native approval capture for the current comment-based workflow.
- +Scoped provider connections plus apply-time rechecks before live mutation.
Current product status: strong technical wedge, still early, and not pretending broad coverage that does not exist yet.
PRICING LOGIC
Priced for governed production control — not seats
Keycli should be priced around protected production targets and governed live changes, because the value is safe execution, approval, and audit — not seat count.
EARLY ACCESS
Team
For AI-native teams starting to put a trust layer in front of real provider changes.
- Base platform fee
- Includes protected targets
- Includes governed live changes
SCALING
Growth
For teams coordinating more providers, environments, and approval paths as agent usage grows.
- More targets and change volume
- More approval workflow depth
- Deeper multi-provider control
CUSTOM
Enterprise
For private deployment, deeper policy requirements, and compliance-heavy environments.
- Private deployment options
- Custom policy and controls
- Higher-touch rollout support
DESIGN PARTNERS
We’re looking for AI-native teams to shape the product with us
If your team is already letting agents touch code, deploys, configs, or secrets, Keycli is being built for exactly that transition point.
- You already use coding agents in real engineering workflows.
- You care about approval, audit, and production trust more than novelty demos.
- You are comfortable shaping an early control-plane product with direct feedback.
Capture needs routing before launch
What happens next
Founder-led and manual on purpose: the goal here is qualified conversations, not fake self-serve. Tell us the workflow you want to govern and we will reply with the right next step.
- We read every request manually.
- If the workflow fit is strong, we follow up with a focused intro call or async thread.
- We keep the pilot narrow and honest rather than forcing a broad onboarding story too early.
FAQ
What teams will ask first
What is Keycli?
Keycli is a trust layer between AI agents and production systems. It turns intent into plans, runs policy and approval checks, applies through provider adapters, and records the outcome.
Is this a secret manager?
No. Secret and config changes are part of the wedge, but the product is about governed production change orchestration rather than static secret storage.
Which providers are supported today?
The current live provider wedge covers Vercel, GitHub Actions, and Render when the right provider connections exist.
Can agents apply changes directly?
They can request changes through Keycli. Risky plans still require approval, and execution is constrained through provider adapters rather than direct prod access.
How do approvals work right now?
The current narrow GitHub-native wedge uses comment-based approval capture, while the hosted control plane also exposes explicit approval endpoints.
DOCS HUB
Use this page to see what is already real, what to run next, and where the truth lives in-repo.
The marketing app now has a proper docs hub instead of a dead-end bridge. It summarizes the real product wedge, points to the canonical demos, and keeps the source-of-truth files visible so the site does not drift.
Real today
- Hosted API with plans, approvals, runs, audit, and scoped auth.
- Live Vercel, GitHub Actions, and Render mutation when valid provider connections exist.
- Mixed-provider execution when every target has supported live adapters.
- GitHub comment approval capture for the current narrow workflow.
Still early
- Full GitHub App install flow and richer PR-native UX.
- Deeper policy and permissioning beyond current scope/recheck guardrails.
- Rollback and drift detection after apply.
Start with the product thesis
If you need the sharpest explanation of what Keycli is and is not, start here before touching marketing copy or roadmap language.
- Trust layer between agents and production
- Normalized plan → policy → approval → apply → audit model
docs/PRODUCT.md
Run the honest demo path
Use the self-hosted control-plane demo first, then graduate to the preview-safe live Vercel flow once you want proof of real mutation.
- demo:hosted:self for the trust model
- demo:vercel:preview for the live provider wedge
docs/DEMO.md
Configure live providers
Provider docs stay explicit about what is live, what is scoped, and what still falls back to simulation.
- Vercel preview-safe mutation guide
- GitHub and Render setup notes for the current supported wedge
docs/VERCEL.md + docs/GITHUB.md + docs/RENDER.md
Keep landing copy honest
The landing specs are still in the repo and remain the fastest way to check message hierarchy, design rules, and anti-drift constraints.
- Overview, copy, wireframe, design system, and pricing notes
- Use them to keep the site aligned with product truth
docs/landing/*
Useful commands
npm run landing:devnpm run landing:typechecknpm run landing:buildnpm run demo:hosted:self