Getting started

Introduction

Vectorbea is a control plane for durable AI workflows. You compose agents, tools and connections on a canvas, run them with full observability, and govern them with human approval gates. Every completed step commits a checkpoint that survives a crash.

What Vectorbea is#

Vectorbea treats a workflow run as a durable, event-sourced entity rather than a process in memory. Waiting, failing, retrying and resuming are normal, cheap operations instead of exceptional ones.

Visual canvas — triggers, agents, LLM calls, tools, branches, parallel fan-outs, approval gates and checkpoints.
Durable runtime — Kotlin workers execute runs from a queue with leases, retries, timeouts and crash recovery.
Observability — every run produces an event timeline, structured logs, per-node token and cost accounting, and failure diagnosis.
Governance — approval gates, risk-tiered tools, read-only connections and a full audit trail.
BYOK — your own provider keys, encrypted with AES-256-GCM and decrypted only in the runner.

The problem it solves#

Real AI automation is long-running. A renewal pipeline waits days for a manager’s decision. A research agent makes fifty LLM and API calls. A migration processes thousands of records. Scripts and basic agent frameworks break down in four specific ways:

Crashes lose everything. A timeout at step 9 of 10 means re-running steps 1–8 — repaying every token and re-firing every side effect.
Side effects are unsafe to retry. Re-running naively sends the same email twice and double-writes to your CRM.
Humans cannot intervene. There is no first-class way to pause for sign-off and continue exactly where it stopped.
Nothing explains itself. When an agent misbehaves you need the exact prompt, tool calls, outputs, latency and cost — not a print statement.

The shape of a run#

A run moves through five states. Only `running` consumes a worker; queued, suspended and failed cost nothing while they wait.

Where it fits#

Vectorbea earns its keep when a workflow is long, expensive, or touches something that cannot be done twice.

Contract renewal pipeline — a nightly trigger pulls upcoming renewals, an agent scores churn risk, high-risk accounts pause at a gate for the account manager, and approved ones get drafted outreach. The run waits days at the gate at zero cost.
Claims and ticket triage — a webhook starts a run, an agent extracts entities and checks policy, low-risk cases auto-resolve and uncertain ones escalate to a gate, resuming with the reviewer’s decision in context.
Monthly analytics — queries, API calls and a composed narrative, each stage checkpointed, so a flaky API at step 4 never repeats the expensive steps 1–3.
Bulk migration — thousands of records in batches with checkpoints between them. A crash at record 8,000 resumes at 8,001.
Deep research agent — iterative loops across search, scraping and synthesis over hours, every iteration an event you can inspect or resume from.
Incident response copilot — an alert starts a run, an agent gathers logs and proposes remediation, and a gate requires on-call approval before anything mutating executes.

Was this page helpful?

What Vectorbea is#

Visual canvas — triggers, agents, LLM calls, tools, branches, parallel fan-outs, approval gates and checkpoints.

Durable runtime — Kotlin workers execute runs from a queue with leases, retries, timeouts and crash recovery.

Observability — every run produces an event timeline, structured logs, per-node token and cost accounting, and failure diagnosis.

Governance — approval gates, risk-tiered tools, read-only connections and a full audit trail.

BYOK — your own provider keys, encrypted with AES-256-GCM and decrypted only in the runner.

The problem it solves#

Crashes lose everything. A timeout at step 9 of 10 means re-running steps 1–8 — repaying every token and re-firing every side effect.

Side effects are unsafe to retry. Re-running naively sends the same email twice and double-writes to your CRM.

Humans cannot intervene. There is no first-class way to pause for sign-off and continue exactly where it stopped.

Nothing explains itself. When an agent misbehaves you need the exact prompt, tool calls, outputs, latency and cost — not a print statement.

Where it fits#

Vectorbea earns its keep when a workflow is long, expensive, or touches something that cannot be done twice.

Contract renewal pipeline — a nightly trigger pulls upcoming renewals, an agent scores churn risk, high-risk accounts pause at a gate for the account manager, and approved ones get drafted outreach. The run waits days at the gate at zero cost.

Claims and ticket triage — a webhook starts a run, an agent extracts entities and checks policy, low-risk cases auto-resolve and uncertain ones escalate to a gate, resuming with the reviewer’s decision in context.

Monthly analytics — queries, API calls and a composed narrative, each stage checkpointed, so a flaky API at step 4 never repeats the expensive steps 1–3.

Bulk migration — thousands of records in batches with checkpoints between them. A crash at record 8,000 resumes at 8,001.

Deep research agent — iterative loops across search, scraping and synthesis over hours, every iteration an event you can inspect or resume from.

Incident response copilot — an alert starts a run, an agent gathers logs and proposes remediation, and a gate requires on-call approval before anything mutating executes.

Was this page helpful?