Start here

Build with Vidbyte

Vidbyte gives you two tools for working with AI agents: vidbyte-sdk, a Python library for building your own agents, and vidbyte-cli, a terminal app that runs research and coding agents for you. This page explains what each one does, what it costs, and where to start.

What is Vidbyte?

An AI agent is a program that uses a language model to work through a task on its own: it reads your request, decides what to do next, uses tools such as web search or file editing, and reports back. Vidbyte makes agents easier to build, run, and pay for.

Build your own AI assistants

The SDK is a free, open-source Python package. You install it with pip, describe an agent in code, and run it on your own computer. By default (vidbyte-sdk 0.4.0 and newer) every model call goes through Vidbyte's model gateway and needs a Vidbyte API key. If you also hold your own provider key, such as OpenAI or Anthropic, the gateway passes it on: the provider bills you for the model and Vidbyte charges a 2% fee. Without one, Vidbyte pays the provider and bills your API balance the provider's rate plus 2%. To skip Vidbyte entirely, set model_mode=TextModelMode.DIRECT and the agent calls the provider straight with your own key, with no Vidbyte key or fee.

What you can build

  • Agents — Give a model a name, instructions, and a provider, then send it a task with run().

  • Tools — Let an agent call your own Python functions, search the web, or read files.

  • Teams of agents — Chain focused agents together so each one handles one part of a larger job.

  • Spending caps — Stop a run once it has spent a dollar amount you choose, with AgentLoopSettings(max_spend_usd=0.50) in vidbyte-sdk 0.4.0 and newer. Most agents support it, and an agent that cannot honor the limit refuses the setting instead of ignoring it.

  • Guardrails — Add retries and safety checks that apply to every run.

  • Evaluations — Test an agent against example tasks and score its answers before you ship it.

What it looks like

agent.py

Python
1from vidbyte import VidbyteSDK 2 3agent = VidbyteSDK().agents.base( 4 name="retry-reviewer", 5 system_prompt="Compare engineering options against the stated constraints.", 6 provider="openai", 7 model_name="gpt-5.5", 8) 9reply = agent.run("Compare backoff with jitter and fixed retries.") 10print(reply.content)
What do I need to run it?

Python 3.11 or newer and a Vidbyte API key with the models:invoke permission (a key created with the default permissions has it) and some API balance, because agents send model calls through Vidbyte by default. Your own provider key, such as OPENAI_API_KEY, is optional: the gateway passes it on to the provider. To run with only a provider key and no Vidbyte account, pass model_mode=TextModelMode.DIRECT. JevAgent always needs a Vidbyte API key, because its Jev calls run through Vidbyte.

What does it cost?

The SDK itself is free and MIT-licensed. In the default mode, each model call carries a 2% Vidbyte fee on top of the model's price: when your own provider key is passed on, the provider bills you for the model and Vidbyte bills only the fee; otherwise Vidbyte bills your API balance the provider's rate plus 2%. With model_mode=TextModelMode.DIRECT, the provider bills you and Vidbyte charges nothing. JevAgent's Jev calls are billed to your API balance at Jev's rate plus a 10% markup on each run.

Run AI agents from your terminal

The CLI is a command-line app named vidbyte-cli. You type one short command in your terminal and an agent does the work. You do not write any code.

What you can do

  • Hosted research — Vidbyte's servers search the web, read the sources, and write up what they found. Ask follow-up questions in the same thread and watch progress while it runs.

  • Local coding agents — Run Codex on your own computer as a small team: a planner, read-only specialists, and one final agent that makes the change. Or keep one agent improving its answer over several rounds.

  • Next-step ideas — Turn a goal into a ranked list of concrete next actions. In the released CLI (0.3.4) it runs on your own Codex or Claude Code, and Vidbyte charges 1 cent per suggestion you ask for.

View more agents

Every agent below runs on your computer and bills model use to your own provider account, so the last column is its only Vidbyte charge. In the released CLI (0.3.4) that includes the suggestion agent, which runs on your own Codex or Claude Code. vidbyte-cli agents list prints these agents in your terminal, along with the bug finder and the infinite memory agent.

Suggestion agent
What it doesDevelops, critiques, and ranks practical next steps for a goal, with handoffs another agent can act on.
Commandagents suggest run
Vidbyte charge$0.01 per requested suggestion
Same-host ensemble
What it doesCreates specialist roles to explore approaches, picks the strongest one, then has one Codex agent build it.
Commandruntime same-host-ensemble
Vidbyte charge$0.02 per run
Persistence agent
What it doesKeeps one Codex session improving its work on a task over repeated rounds.
Commandruntime persistence
Vidbyte charge$0.02 per run
Task board
What it doesRuns an ordered list of tasks, with a fresh Codex agent for each task and optional checkpoints.
Commandruntime task-board run
Vidbyte charge$0.02 per run
Stages
What it doesRuns stages you define, one after another or in parallel, each with a fresh Codex agent.
Commandruntime stages run
Vidbyte charge$0.01 per run

What it looks like

Start a research thread

bash
vidbyte-cli research start "Compare three ways to reduce latency in an LLM tool-calling workflow. Cite sources."
What do I need to run it?

Python 3.11 or newer, a Vidbyte account, and a Vidbyte API key. Local coding agents also need the codex command on your PATH, which pip does not install. Next-step ideas in the released CLI (0.3.4) need a key with the runtime:write permission (an admin or full-access key also works), balance for the suggestions you ask for, and Codex or Claude Code installed and signed in on your computer.

What does it cost?

Hosted research is paid from your Vidbyte API balance. Local coding agents charge a fee per run from that balance: $0.01 for stages, $0.02 for the same-host ensemble, persistence agent, and task board. The adversarial team is coming soon; its listed fee is $0.25 and nothing is charged until it launches. The model work is billed to your own OpenAI account or Codex subscription. For next-step ideas, the released CLI (0.3.4) charges 1 cent per suggestion you ask for before the run starts, OpenAI or Anthropic bills the model use to your own account through Codex or Claude Code, and a dry run is free.

Your API key and balance

The CLI's paid features need two things from your account: a Vidbyte API key, which tells Vidbyte who is making the request, and API balance, which pays for the work. SDK agents need both too in their default mode, which sends model calls through Vidbyte; an SDK agent set to model_mode=TextModelMode.DIRECT needs neither.

Calling the HTTP API directly

The CLI's hosted features use Vidbyte's public HTTP API. If you build your own client, send requests to this base URL with your API key. The API also creates learning resources such as roadmaps, quizzes, and projects.

Public API host

https://api.vidbyte.pro

Next steps