Skip to main content

How to run a first-line support agent with Talon

Run a first-line support agent that summarizes tickets and suggests replies, with PII and cost controls. You can implement it as a native Talon agent (webhook or cron trigger) or by pointing an existing support bot at the LLM API gateway.

Option A: Native Talon agent​

Use a Talon agent with a webhook trigger (e.g. from Zendesk, Freshdesk) or a cron job. Policy enforces cost limits and PII handling; every run is recorded as evidence.

1. Policy snippet​

In agent.talon.yaml:

agent:
  name: "support-agent"
  description: "First-line support: summarize tickets and suggest replies"

policies:
  cost_limits:
    daily: 20.00
    monthly: 400.00

# PII: block or redact before LLM sees it
# (exact key depends on your policy schema; often default_pii_action or capabilities)

Optional: enable plan review for high-risk suggestions (compliance.human_oversight: "on-demand"). See Policy cookbook and Agent planning.

2. Trigger​

Webhook: Expose POST /v1/triggers/support-incoming and point your ticketing system at it. In .talon.yaml:

triggers:
  webhooks:
    - name: support-incoming
      source: zendesk
      prompt_template: "Summarize this ticket and suggest a reply: {{.payload.description}}"
      require_approval: false

Cron: Run on a schedule, e.g. poll a queue and run talon run "Process next support ticket..." from a script.

3. Serve and run​

talon serve --port 8080

Evidence is stored per run. Use talon audit list or GET /v1/evidence filtered by agent_id or time to answer "what did the support agent see on date X?"

Option B: Existing support bot → gateway​

If you already have a Python/Node/etc. bot that calls the OpenAI (or other) API:

  1. Add a gateway caller for the support bot (e.g. support-bot) with policy_overrides.max_daily_cost and pii_action: "block" or "redact".
  2. Point the bot at the Talon gateway base URL and use the caller API key instead of the real provider key. Same pattern as How to add compliance to your Slack bot.
  3. Ticket content (often containing PII) is scanned by Talon; policy can block or redact before the request reaches the LLM.

Evidence is recorded per request; costs appear in GET /v1/costs for the tenant. Filter evidence by caller name to see support-bot usage.

Evidence and “what did the agent see?”​

  • Native: talon audit list, talon audit show <evidence-id>, or API GET /v1/evidence with optional filters. Evidence includes correlation_id, tenant_id, agent_id, timestamp, policy decision, cost.
  • Gateway: Same evidence store; gateway requests are recorded with caller identity. Use tenant and date (and caller if exposed in evidence) to scope exports for auditors. See How to export evidence for auditors.

You're done​

You now have a first-line support agent (native or via gateway) with PII and cost controls. Talon is recording every run and applying your policy.

Next steps:

I want to…Doc
Route another bot through the gatewayAdd Talon to your existing app, Slack bot
Export evidence for auditorsHow to export evidence for auditors
Add plan review for high-risk suggestionsAgent planning, Policy cookbook
Cap cost for this agent or callerHow to cap daily spend per team or application