Introducing Vektra Agents 2.0: multi-step reasoning for every team

Today we're releasing Vektra Agents 2.0, the biggest update since we launched. Agents can now break a request into steps, use several tools in sequence and adapt when something doesn't go as planned.

What's new in 2.0

Until now, most agents — ours included — were great at one thing at a time: answer this question, update that record. Real work rarely looks like that. A single customer email can require checking an order, reading a policy, issuing a refund and updating the CRM.

• Planning: agents draft a plan before acting and show it to you.
• Tool chaining: outputs from one step flow into the next automatically.
• Self-correction: when a step fails, the agent retries with a different approach.

“Agents 2.0 handles the messy, multi-system tickets we used to escalate. It feels like hiring a senior teammate.” — Maya Chen, Quantix

How multi-step reasoning works

Each run starts with a planning phase. The agent reads the request, the available tools and your instructions, then proposes a sequence of steps. You can inspect every plan in the run trace.

plan:
  1. lookup_order(id: "4821")
  2. read_policy(topic: "damaged items")
  3. issue_refund(amount: 184.00)  # requires approval
  4. reply_customer(template: "refund_confirmed")

Guardrails & approvals

More autonomy demands better guardrails. In 2.0 you can define approval rules per action and per amount, set spending limits, and restrict which tools each agent can access. When an approval is needed, the agent pauses and explains its reasoning in one sentence.

Getting started

Agents 2.0 is available today on all plans. Existing agents are upgraded automatically, and you can opt out per agent from the settings page. Read the migration guide or start a free trial to try it yourself.