The product
One platform. Four surfaces.
Chat and Sandboxes are what your teams use. Console and Tracing are what you control them with. All four run on your infrastructure.
Chat
Conversation is the interface.
Your team delegates work in a thread and sees when it's done. Nothing new to learn.
chat.agyn.dev
Q3 supplier spend variance
Dana Kerr, @finance-analyst · 2 minutes ago
DK
Dana Kerr
Which suppliers drove the Q3 overspend? Chart it against budget.
@finance-analystView trace
Four suppliers account for 82% of the variance. Northwind is the outlier at +$412K against a $180K budget line.
Q3 spend vs budget ($K)
ActualBudget
Northwind
Cortek
Halberd
Vantage
Message @finance-analyst…
Sandboxes
Engineers get a shell in the agent's runtime.
Same image, same secrets, same limits — driven by hand. Find out what an agent can and cannot reach before it runs unattended.
sandboxes.agyn.dev/brave-otter
agyn/Sandboxes/brave-otter
MR
RunningEnvironment backend-devStops in 42mExpires in 6h
Ports3000http://brave-otter.acme.agyn:30008080http://brave-otter.acme.agyn:8080Expose a port
buildserver3
➜ /workspace/api pnpm dev
> api@0.4.2 dev
> tsx watch src/server.ts
> tsx watch src/server.ts
ready — listening on http://0.0.0.0:3000
warn — DATABASE_URL resolved from secret prod-readonly
✓ connected to postgres.corp:5432 via egress rule
GET /health 200 4ms
GET /api/orders?q=q3 200 118ms
GET /api/orders?q=q3 200 118ms
➜ /workspace/api
Console
One place to set the rules.
Every token, CPU-hour, and gigabyte lands on the agent that used it. Same page decides who can use which model, and what it's allowed to reach.
console.agyn.dev/operations/usage
Usageacme
AV
Last 24 hoursLast 7 daysLast 30 days
Input tokens
84.2M
Cached
61.5M
of 84.2M input
Output tokens
9.9M
Successful
12,408
Failed
37
Tokens over time
CachedFresh inputOutput
20M
10M
0
MonTueWedThuFriSatSun
Top consumers
@finance-analyst32.1M
@support-triage24.8M
@sre-agent15.2M
@compliance-agent9.4M
Dana Kerr5.8M
Tracing
Every run is a record you can read.
Every LLM call, tool execution, and context decision — in order, with the prompt that produced each response. Terminate a run that's misbehaving.
tracing.agyn.dev/runs/wl-8f21c4
agyn/acme/wl-8f21c4
Running@finance-analyst · 18.2s
Input 84,210Cached 61,904Output 3,118Reasoning 1,442Terminate
LLM Callanthropic / claude-opus-5
Stop reason
tool_use
Temperature
0.2
Duration
4.0s
Tool calls
1
Context — 34 items
systemYou are a financial analysis agent…2.1 KB
tool_resultpostgres_query → 412 rowsNEW18.4 KB
summaryPrior context condensed at 14:02:201.2 KB
Response
Four suppliers account for 82% of the variance. Charting actual against budget for the top four now.
See it on your own cluster.
Thirty minutes, your infrastructure, your stack. Or start it on your laptop right now and skip the call.