CosavuCosavu

← Products/VexaAgent

Agents that don't hallucinate.
Powered by your data.

VexaAgent is a production-grade agent runtime built on Vexa-1, ContextAPI, and DataAPI. Tool use, grounded reasoning, and token-aware planning — out of the box.

0%
ungrounded answers
1.4s
median agent latency
30+
built-in tools

The full stack, orchestrated

Three APIs.
One agent loop.

VexaAgent orchestrates ContextAPI for prompt optimisation, DataAPI for grounded retrieval, and Vexa-1 for reasoning — every loop, every turn.

You write the goal. The agent plans, retrieves, composes, and verifies — without you wiring up the orchestration.

agent loop
> AGENT INVOKED
goal: "draft a customer reply citing
the refund policy"
PLAN ─────────────────────────────
├─ retrieve refund policy
├─ retrieve customer thread
└─ draft response
STEP 1/3 thinking ········· |

How it works

One loop.
Four primitives.

01

PLAN

decompose goal

02

RETRIEVE

DataAPI grounding

03

COMPOSE

ContextAPI + Vexa-1

04

VERIFY

self-check & cite

↻ Loop continues until goal is satisfied or budget is exhausted. State persists between iterations.

Capabilities

What VexaAgent does

01

Multi-step reasoning

Plans tasks into discrete steps, executes them in order, and replans on failure. No hand-rolled state machines required.

02

Native tool use

Calls ContextAPI, DataAPI, and your custom tools natively. Tool schemas auto-validated; failed calls are retried with adjusted inputs.

03

Grounded by default

Every response is backed by DataAPI retrieval. Hallucinations are blocked by source-citation enforcement at the agent level.

04

Token-aware

All prompts pass through ContextAPI before reaching Vexa-1. Long-running agents stay within budget without truncating context.

05

Persistent memory

Agent state, tool outputs, and intermediate reasoning persist across turns and sessions — keyed per user, per workflow.

06

Streaming & async

Token-by-token streaming for chat. Background mode for long-running workflows with webhook callbacks on completion.

Use cases

Built for the agents
you actually ship.

Customer ops, internal copilots, research assistants, ticket triage, code review — workflows where hallucinations cost real money.

support

Customer support copilot

Drafts responses grounded in policy docs and prior tickets — never invents a refund window.

ops

Ticket triage & routing

Reads incoming tickets, retrieves runbooks, classifies severity, routes to the right team.

research

Internal research assistant

Synthesises across your knowledge base with full citations. No more 'I think it's in Notion somewhere.'

code

Code review & PR summaries

Pulls relevant code context, reviews against your style guide, posts grounded review comments.

Early access

VexaAgent is in private beta.

Join the early access list. We're onboarding teams shipping agents to production.