Agents

Agents that run your operations, under contract.

A G8 agent is an artificial colleague: a described process, made executable, that does governed work on your behalf. It acts only within an contract, and it stops for a human whenever the stakes are real.

Contract-bound Human on risk Fully audited

01 / What an agent can do

Governed work, not a free-roaming bot.

Each capability is something G8 actually does today — bounded execution, human oversight, and a record of everything that happened.

Delegate a whole process

Describe a process in plain language and G8 compiles it into a Blueprint — the steps, the tools touched, the risk of each. The agent is that process, made executable.

Natural language → Blueprint

Act only within a contract

Every agent runs under an Automation Contract: an explicit allow-list, a block-list, and the steps that require approval. It cannot improvise past its mandate.

Allow · block · approval-required

Pause for a human on risk

Anything irreversible — spending money, messaging outsiders, deleting data — stops at an approval gate. The run waits for a decision, then resumes from where it paused.

Human-in-the-loop

Orchestrate many in parallel

Run a fleet of flows at once, each triggered by a schedule, a webhook, or a manual start. They execute independently while sharing the same governance.

Schedule · webhook · manual

Reach the systems you use

Agents act through real connectors — HTTP, Telegram, Email, and Slack — so a step can call an API, message a channel, or send mail as part of the run.

HTTP · Telegram · Email · Slack

Leave a full audit trail

Every step, approval, retry, and output is recorded against the run. You can trace exactly what an agent did, when, and on whose sign-off.

Step-level traceability

02 / How an agent runs

The same five steps, every time.

From a description to a running, monitored flow — the lifecycle every agent follows.

  1. 01

    Describe

    Hand over the process

    Write the work down the way you'd explain it to a colleague. G8 compiles it into a Blueprint with steps and risk levels.

  2. 02

    Contract

    Bound what it may do

    Set the allow-list, the block-list, and the approval-required steps. The contract is the agent's mandate.

  3. 03

    Simulate

    Dry-run with no effects

    Walk every step and read the expected output before anything is sent, posted, or written. Cheap to check, cheap to fix.

  4. 04

    Run

    Execute, pause on risk

    Live, the agent runs on its trigger and pauses at every approval gate. Approved runs resume; rejected runs stop cleanly.

  5. 05

    Monitor

    Watch and trace

    See run volume, success rate, errors, and what's waiting on a human — and drill into any run step by step.

Get started

Hire your first agent.

Describe a process, contract it, simulate it, and let it run — under your terms.

Delegate a process