Run AI agents safely
across your enterprise
Deploy and orchestrate agent teams company-wide with enterprise-grade security
Open-sourceKubernetes-nativeClaude · Codex compatible
$ git clone https://github.com/agynio/bootstrap
$ cd bootstrap
$ ./apply.sh -y01
Multi-envPrivate networksVPN · VPCInstant rollback
Deploy into your VPC in minutes.
Ship agents into any environment or private network — reaching internal services behind VPNs, VPCs, and firewalls.
02
Least privilegeStatic policiesPolicy agentSecrets vaultPrompt-injection defense
Secure every action before it runs.
Every agent runs under least privilege. Static policies and a policy agent inspect every tool call. Secrets stay hidden from the model — defending against prompt injection and sensitive data leaks.
03
Per-agent trackingBudget limitsUsage alertsCost attribution
Budgets and alerts for every agent.
Track spend across agents, teams, and workflows. Set budgets, get alerts, and keep AI costs under control.
04
Team sharingAccess controlRole-basedAudit logs
Role-based access across teams.
Give the right employees access to the right agents, share them safely across teams, and keep usage governed as adoption grows.
05
AgentsSandboxesMCPs & skillsGitOps
Define agents and policies in git.
Declare agents, sandboxes, tools, MCPs, skills, and prompts in code. Agents run consistently across every team and environment.
Local development
The problem
Claude — Data Analysis Agent
JK
You
Analyze Q1 sales data from our production database and create a summary report.
C
Claude
I'll connect to your production database and analyze the Q1 sales data.Unfortunately, I cannot access your production database from my laptop. It's only accessible from inside your corporate network.
Connecting to postgres://internal.corporate.network:5432...
⚠ Connection failed
Error: Cannot reach internal.corporate.network
This resource is only accessible from corporate network
My local environment cannot access internal services
This resource is only accessible from corporate network
My local environment cannot access internal services
❓ How do I give Claude access to corporate network?
❓ How do I deploy this so my whole team can use it?
❓ How do I scale this across the company?
❓ How do I deploy this so my whole team can use it?
❓ How do I scale this across the company?
agyn deploy
Enterprise deployment
Solved
Data Analysis Agent
Deployed to Corporate Network
internal.corporate.network
internal.corporate.network
Live
99.9%
Uptime
18
Team users
142
Requests
1.2s
Deploy time
acme-corp/⚡ Platform Team/Security
Agent scopes
Each agent gets only what it needs
Code Reviewer
codex-gpt-5
read reposcomment on PRssend emailexternal network
Support Agent
claude-haiku-4.5
send emailread ticketsread reposread database
Data Analyst
claude-opus-4.6
read databasewrite reportswrite databasesend email
Researcher
gemini-2.5-pro
browse webwrite summariesinternal accesssend email
Policy gate
Every action reviewed before it runs
247
Allowed today
18
Blocked today
3
Escalated
Policy Agent
Reviews every action before execution
Active
Active policies
Least privilege
Actions outside agent scope are dropped
Planner / Executor split
Planning agents cannot directly execute
Sanitise feedback
Strip injected instructions from responses
Human escalation
Pause on ambiguous or high-risk actions
Live outcomes
What the policy gate is deciding right now
Allownow
Code Reviewer
Read repo and post review comment on PR #482
Within scope · no external calls
Block2m ago
Code Reviewer
Email review summary to partner@external.com
send email not in scope · dropped
Allow3m ago
Support Agent
Reply to customer ticket #8821 via email
Within scope · internal domain only
Block5m ago
Support Agent
Query user records from internal database
read database not in scope · dropped
Block8m ago
Researcher
Follow instructions embedded in web page
untrusted source · instruction stripped · escalated
Allow11m ago
Data Analyst
Read Q1 sales data and write report
Within scope · read-only enforced
acme-corp/Cost & Usage
Overview
By agent
By team
Budgets
Month-to-date
$4,445
↑ 12% vs last month
Budget remaining
$2,655
26 days left · $7,100 limit
Total requests
184,420
↑ 8% vs last month
Avg cost / request
$0.024
↓ 4% more efficient
Daily spendApr 1 – 17
Code Reviewer
Data Analyst
Researcher
Team budgetsmonthly limits
Platform Team74% used
$1,847 spent$2,500 limit
Engineering41% used
$820 spent$2,000 limit
Data Science58% used
$1,160 spent$2,000 limit
Customer Success103% — over
$618 spent$600 limit
Customer Success exceeded its monthly budget by $18. Requests are rate-limited. Increase limit →
acme-corp/Platform Team/Access
Team agents4 agents
Code Reviewer
Platform Team
8
Support Agent
Customer Success
12
Data Analyst
Data Science
6
Researcher
Marketing
9
Code Reviewer — Access
Platform Team · 8 members with access
Members
Permissions
Activity
BK
Ben Kovich
ben@acme.com
SR
Sara Reyes
sara@acme.com
JT
James Torres
james@acme.com
LP
Lena Park
lena@acme.com
DM
Dev Mehta
dev@acme.com
AW
Alex Wu
alex@acme.com
MJ
Maya Johnson
maya@acme.com
TC
Tyler Chen
tyler@acme.com
6with access
2no access
Last updated 2 min ago
📄sandbox.tf
📄agents.tf
📄mcps.tf
📄policy.tf
Explorer
▾platform-team
📄sandbox.tf
📄agents.tf
📄mcps.tf
📄policy.tf
📄skills.tf
▾prompts
📝reviewer.md
📝analyst.md
📝researcher.md
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
# Sandbox — isolated execution environment
resource "agyn_sandbox" "default" {
network = "corporate-vpn"
policy = agyn_policy.least_privilege.id
}
# Agents — each scoped to its sandbox
resource "agyn_agent" "reviewer" {
name = "Code Reviewer"
model = "codex-gpt-5"
instructions = file("prompts/reviewer.md")
sandbox = agyn_sandbox.default.id
skills = [agyn_skill.code_review.id]
mcps = [agyn_mcp.github.id]
}
resource "agyn_agent" "analyst" {
name = "Data Analyst"
model = "claude-opus-4.6"
instructions = file("prompts/analyst.md")
sandbox = agyn_sandbox.default.id
skills = [agyn_skill.sql_analysis.id]
mcps = [agyn_mcp.database.id]
}
resource "agyn_agent" "researcher" {
name = "Researcher"
model = "gemini-2.5-pro"
instructions = file("prompts/researcher.md")
sandbox = agyn_sandbox.default.id
mcps = [agyn_mcp.browser.id]
}
Terminal
Problems
Output
$ terraform apply
Planning changes for platform-team…
+ agyn_sandbox.default · agyn_policy.least_privilege
+ agyn_mcp.github · agyn_mcp.database · agyn_mcp.browser
+ agyn_skill.code_review · agyn_skill.sql_analysis
+ agyn_agent.reviewer · agyn_agent.analyst · agyn_agent.researcher
Apply complete! Resources: 10 added, 0 changed, 0 destroyed.
Ready to run AI agents in your enterprise?
See how agyn deploys secure, governed agent teams across your organization — live, with your stack.
Book a demo