Support Agent
LMN-2041Answers tickets, looks up orders and drafts replies.
tickets.readorders.getpolicy.search
Waits for a person: Refunds and credits
Agents
Each agent has one job, a catalogue ID and a set of tools. You watch the fleet, approve what matters, and read the trail of anything it did.
Catalogue
A job title and a catalogue ID, never a human name or face. These five are examples of what a pilot can start with.
Answers tickets, looks up orders and drafts replies.
tickets.readorders.getpolicy.search
Waits for a person: Refunds and credits
Handles invoices, failed payments and plan changes.
invoices.getpayments.retryplans.change
Waits for a person: Any refund or plan downgrade
Reads your data and writes the weekly summary.
warehouse.querydocs.searchreport.write
Waits for a person: Sending a report outside the team
Reviews pull requests and runs the tests.
repo.readtests.runpr.comment
Waits for a person: Merging or pushing to main
Sets up new accounts and sends the welcome sequence.
accounts.createemail.draftcrm.update
Waits for a person: The first email to each account
What every agent does
No step runs before the plan is written down. You can approve it, change it or stop it.
Each step shows what it did, the tool it called and its time. Each session shows steps · time · cost.
The trail of every session is kept in an append-only log you can export.
Refunds, merges and anything else you mark irreversible wait for a person.
When a case needs judgement, the agent hands it to a named person with the context attached.
A live timer runs while an agent works, next to a Stop button that ends the session at once.
The console
The selected agent shows its approved plan, each traced step, the approval gate, its receipt and a live timer. Press Stop and it stops.
| State | Agent | Task | Receipt |
|---|---|---|---|
| Working | Refund request, order 88213 | 6 steps · 0:41 · $0.08 | |
| Waiting on you | Approve refund of $48.00 | 9 steps · 1:12 · $0.11 | |
| Handed off | Damaged item, photo review | 4 steps · 0:22 · $0.03 | |
| Done | Weekly churn summary | 14 steps · 3:05 · $0.21 | |
| Failed | Pull request 418, tests timed out | 7 steps · 2:40 · $0.09 | |
| Queued | Welcome sequence, 12 accounts | not started |
Sample data throughout. Example
Open the live Inferstack ConsoleUnder the hood
Long agent sessions resend the same context again and again. The session layer is built so they don’t pay for it twice.
Requests from one agent session go to the GPU that already holds its KV cache, so old context isn’t computed again.
KV cacheSession routing
Stable parts go first and new turns go last: system prompt, tools, repo map, then turns. One changed token early in the prompt breaks the cache for everything after it.
Prefix cachingPrompt layout
Three policies: cost, balanced and quality. Under balanced, chat and extraction go to a small model; code and reasoning go to a large one.
Step routingSmall + large models
The session lives in a log that no single model owns, so a new model can pick it up where the last one stopped.
Session logModel handoff
Stop at a clean turn boundary.
Fold old turns into the session log, keeping the plan, decisions and open files.
Write the log in the new model’s chat template and tool format.
One prefill fills the new model’s cache.
A short probe confirms the new model knows the plan.
The agent continues from the same turn.
A pilot is one agent, set up with us, on your own work. You keep the receipts.