T3 Code explained: the client controls work on another machine
T3 Code performance and cost: measure the connection you actually use
Separate UI transport, server startup, provider usage and repository work
What you will learn
- Measure remote responsiveness
- Measure the agent separately
- Account for operational work
Before you start
- A disposable project on an environment host
- One authenticated supported provider
Turn a disposable trial into evidence before connecting important repositories
Key takeaways
- Client transport and provider execution are different costs.
- Subscription design is not a benchmark result.
- Accepted changes matter more than message volume.
Measure remote responsiveness
Track connection establishment, reconnect, time to first streamed response and UI update latency over the route you chose. LAN, SSH forwarding and T3 Connect may differ; keep the same task and host when comparing.
The architecture uses subscriptions to avoid sending every thread’s full history to a viewer. That design suggests a measurement, but this series has no production latency figures and makes no speed claim.
Measure the agent separately
The provider runs on the environment host, so model usage, terminal CPU and Git operations belong to that machine. Record provider version, permission mode and the task before comparing runs.
A desktop app may bundle a server, while hosted web is only a client. Do not put the hosted UI’s hosting cost in the same bucket as the machine running agents.
Account for operational work
Background service, HTTPS or private networking, backups, access revocation and updates each add time. A remote workflow that saves travel but complicates credential recovery may not save overall effort.
Use an outcome measure such as accepted code changes or time to review, not just message count. No benchmark or billing experiment was run for this series.
Decision guide
| Criterion | Option A | Option B |
|---|---|---|
| Best when | You need predictable behavior and easy auditing | You need adaptive optimization and have reliable telemetry |
| Main risk | May leave performance on the table | Can become difficult to explain or debug |
Implementation steps
- 1
Measure connect, stream and reconnect delays separately.
- 2
Track provider cost and host resource use.
- 3
Include access, update and review effort.
Copy-ready example
route,host,provider,connect_ms,first_response_ms,reconnect_ms,accepted_change,notes
,,,,,,,not-runFrequently asked questions
Does hosted web run models for free?
No. The environment host still runs an authenticated provider with its own usage terms.
Which connection route is fastest?
This series did not benchmark routes; measure them on the same host and task.
Sources
- T3 Code / docs/internals/overview.mdSource checked 2026-10-04
- T3 Code / docs/internals/remote.mdSource checked 2026-10-04
- T3 Code / docs/user/remote-access.mdSource checked 2026-10-04
- T3 Code / docs/user/background-service.mdSource checked 2026-10-04