ABot user manual
Follow the setup screens, learn the workspace, and build your first useful agent workflow. Each chapter explains the choices as you make them.
On this page
This is a practical manual for the published ABot Runtime. Start with a working installation, follow the same controls you see in the Web UI, and check a small result before giving the agent more access.
You do not need to configure everything at once. A model connection and an ordinary conversation are enough to begin. Memory, computer access, SPARK and automation can come later.
First time using ABot?
- Install ABot. Prepare Node.js, create your
my-abotdirectory and open the local Web UI. - Complete onboarding, screen by screen. Choose a provider, fill in the connection and model fields, understand the context-window value, then decide what to enable in Embedding, Plugins and Computer.
- Find your way around the Web UI. Learn where conversations, models, permissions, attachments, activity, configuration and system controls live.
- Finish a first task. Check the model connection, fetch one public page, and inspect what actually happened.
You should finish this path with one connected model, a conversation you can return to, and a clear understanding of which actions need approval.
Choose what to learn next
| I want to… | Follow this chapter |
|---|---|
| Make a model behave more reliably | Calibrate a model, step by step: establish a baseline, find the failing step, change one setting, then retest. |
| Understand EA versus SUPER | Choose a model and execution route: direct execution versus focused responsibilities, using the same engine. |
| Know what each JSON setting means | Configuration files, then the detailed main, model and request-runner references: ownership, defaults, precedence and examples. |
| Build a repeatable process | Build a Studio workflow: configure a template, connect inputs, publish, run and inspect outputs. |
| Try optional activity learning | Use SPARK: complete consent, select models, set limits, review learning and understand proactive chat. |
| Give the agent tools without unrestricted access | Projects and permissions and Plugins. |
| Carry useful information across conversations | Set up memory: choose and test an embedding model, then inspect saved records. |
| Run one request later | Schedule a conversation request: choose timing, model and permissions, then inspect a run. |
Work through examples safely
The guides use non-sensitive sample content. Replace placeholder model IDs and paths with values that exist in your installation. JSON blocks say whether they are complete examples or fragments to merge into an existing file.
A successful tool call is not proof of a correct final result. Open the produced file, compare its content with the source, and review errors before repeating a task unattended.
Running ABot locally does not make hosted model calls or external tools local. Generation, embeddings and SPARK may use different providers. Check each connection before sending sensitive content. See the privacy notice and the relevant chapter’s data-handling notes.
Keep learning without resetting your setup
Use Configuration files to understand which file owns a setting. The calibration field guide contains longer worked examples; it complements the guided calibration exercise.
When something fails, use Troubleshooting to identify the layer rather than reinstalling everything. Update your installation explains how to keep configuration and state across package upgrades.
If you are embedding the engine in your own product, Use ABot as a library introduces the public integration surface and links to the detailed contracts.