v0.3.0 · self-serve open

The control plane
for your AI agents.

Sansin sits in front of every agent you run — DevOps, coding, support, your own — and scores every tool call before it executes. Safety that learns. Governance that compounds. Start with your on-call agent.

See a real decision
Read-only by default. Open-source core. One email when a spot opens.
app.sansin.ai/agent · pending
Awaiting approval INC-4127 · 2s ago
Restart prod-api with larger pool
Connection pool exhausted. 18 prior incidents resolved by raising pool_size + restart. Risky during business hours. 
# config/api.yaml
- pool_size: 20
+ pool_size: 50
# action: kubectl rollout restart deploy/prod-api
Risk
0.58
Confidence
0.71
Prior
β(8,12)
Every write
scored against your
per-tool, per-tenant prior
Pre-trained priors for the tools your agents already call

Illustrative trace

The on-call loop you already run. Without you in it.

You used to open Grafana, paste logs into Claude, ask what's wrong, then type the kubectl command yourself. Sansin runs that loop on its own and stops at the write. The same gate governs coding and support agents the moment you wire them in.

1
Gate between every agent and your infra.
1
Click from you. The approval.
sansin · INC-4127 · live trace
0.0s
15:24:03 observed alert · prod-api CPU 89% sustained 3m
15:24:04 observed read_logs api-gateway → upstream timeout
15:24:05 observed read_logs order-service → DB pool exhausted
15:24:06 proposed update_config pool_size=50, restart=true
15:24:06 gate risk 0.58 · conf 0.71 → PENDING
15:24:34 observed sarah@: "try scale_pods first"
15:24:35 proposed scale_pods replicas=6
15:24:35 gate risk 0.18 · conf 0.82 → ALLOW
15:24:36 exec kubectl scale deploy order-service --replicas=6
15:28:55 resolved p99 < SLO · INC-4127 resolved · 4m 19s

How the gate works

Three things sit between every agent and your infra.

01 · Score

Risk-rated, every call.

Each tool call gets a Thompson Sampling score against your tenant's prior for that specific tool. Low risk: auto-approved. High risk: parked.

scope · per tool, per tenant
02 · Gate

You decide on the writes.

Reads run free. Writes that exceed your threshold queue in the dashboard with the diff, the rationale, and one-click approve or reject with reason.

verdicts · allow · block · pending
03 · Learn

Your overrides train the prior.

Every override updates a per-tool Beta distribution. The same prior is shared across every agent in the tenant, so lessons learned on one agent sharpen decisions on the next.

prior · Beta(α, β) per tool

The dashboard

One queue. Every agent. Every pending write.

Pending writes on the right. The full agent transcript on the left. One shared audit log across DevOps, coding, and support — not three tabs.

app.sansin.ai/agent
DevOps agent
Connected · session sess_a8f2e1 · ● running
last 24h
Decisions
347
Auto-allowed
312
Parked
28
Avg risk
0.21

Event feed

15:24:03alertprod-api CPU 89% sustained 3m
15:24:05tool_useread_logs api-gateway
15:24:05tool_useread_logs order-service
15:24:06proposeupdate_config pool_size=50, restart
15:24:06gatePENDING risk 0.58 conf 0.71
15:24:34overridesarah@: try scale_pods first
15:24:35proposescale_pods replicas=6
15:24:35executeALLOW risk 0.18 conf 0.82

Pending approvals · 2

drain_node risk 0.72
node = worker-07
grace = 60s
rotate_secret risk 0.81
secret = stripe_live_key

Tools the agent already speaks

Wires into the things you already page on.

Sansin uses MCP, so any tool server you've already wired to Claude Code works the same day.

Kubernetes
AWS
Datadog
PagerDuty
GitHub
Terraform

Plus any MCP-compatible tool server. See the integrations catalog →

Pricing

Start free. Pay per decision when you scale.

Hobby

Personal projects, side agents, and kicking the tires.

$0forever

No credit card. 1 team member.

  • 25 decisions / month
  • 2 MCP servers
  • Full dashboard + approval flow
  • Open-source core on GitHub
Join the waitlist
Recommended

Pro

Teams running one or more agents in production.

$29/mo + $0.05 / decision

14-day trial, no card required.

  • Unlimited decisions, metered at $0.05 each
  • 10 team members · all MCP servers
  • Slack + Teams approval routing
  • 90-day audit trail
Join the waitlist

Enterprise (SSO, self-host, volume pricing) and the usage calculator on the pricing page →

One control plane.
Every agent you run.

We're onboarding teams one at a time while the gate earns its priors. Ask for a seat and we'll reach out when yours is ready.

Join the waitlist Book a walkthrough

Read-only by default · open-source core · self-host available