Agent integrations
Every agent framework speaks OpenAI. So do we.
OpenCode, Cline, Kilo Code, Hermes Agent, OpenClaw — they all accept an OpenAI-compatible endpoint. Point one at Run BiOS and your agent runs on the same API, whether it is serverless or your own dedicated deployment.
from openai import OpenAI
client = OpenAI(
base_url="https://api.runbios.ai/v1",
api_key="sk-bios-...",
)
resp = client.chat.completions.create(
model="deepseek-v4-flash", # any served model id
messages=[{"role": "user", "content": "Summarize my inbox"}],
)Three values — base URL, API key, model id. Every framework below takes the same three, just in its own config.
The setup
Three values. That is the whole integration.
Agent frameworks differ in memory, tools, and channels — but they all speak the same wire protocol. Swap the endpoint, keep everything else.
The OpenAI-compatible endpoint. Same path shape as api.openai.com — the framework does not know the difference.
Your Run BiOS key from the platform console. One key works for every served model.
Pick from the model library — or the id of your own fine-tuned checkpoint on a dedicated endpoint.
Frameworks
Where the three values go
Each framework keeps its own config format — the values never change. Exact file paths vary between versions; the pattern below is the one every version accepts.
Coding agents
OpenCode
An open-source terminal coding agent — reads your codebase, plans and executes changes, runs tests, and iterates from natural language.
Provider config (opencode.json) — add your API key and base URL, pick any served model id.
Cline
The coding agent that lives in your IDE and plans, edits, and runs your code.
API provider settings in the extension — base URL, key, model id.
Kilo Code
An open-source coding agent for VS Code, JetBrains, and the CLI — bring your own keys, pick from hundreds of models.
Provider settings — add your API key and base URL, then pick any served model id.
Pi
A minimal terminal coding harness you extend with TypeScript packages — read, write, edit, and bash tools out of the box.
Provider config — authenticate with your API key, select the model with /model.
Aider
A terminal pair-programmer that edits git repos with you.
Environment variables — OPENAI_API_BASE and OPENAI_API_KEY.
Continue
An open-source IDE assistant with custom models and context.
config.json — apiBase and apiKey on the model entry.
Personal & messaging agents
Hermes Agent
A self-improving agent with persistent memory, skills, and messaging gateways — Telegram, Discord, Slack, WhatsApp.
Provider config — accepts any OpenAI-compatible endpoint; switch models with `hermes model`.
OpenClaw
A self-hosted gateway that connects WhatsApp, Telegram, Signal, iMessage and more to your AI agent.
Model provider config in the gateway — hosted and local providers both supported.
MIRA
A self-hosted personal agent that reaches you on Signal, Telegram, Discord, WhatsApp, Slack, and email.
Provider config — OpenAI-compatible servers supported; point it at the Run BiOS endpoint.
LibreChat
A self-hosted chat UI that speaks to any OpenAI-compatible API.
Endpoint and key in the UI — no code changes.
Not listed? If it accepts an OpenAI-compatible endpoint, the same three values apply. Per-framework walkthroughs are being verified and will be linked here as they pass.
Dedicated deployment
Your weights, your endpoint, the same three lines.
Fine-tune a model on dedicated GPUs, then serve it on a dedicated endpoint behind the same OpenAI-compatible API. The agent config does not change — only the model id does, and the weights behind it are yours.
- The checkpoint is a file you own and can export
- Dedicated hardware, billed per second of GPU time
- The same base_url and api_key as serverless
Illustrative — the agent config is identical to serverless.
Why Run BiOS
Agents are the hardest workload. Run them on the platform built for it.
Agents burn tokens. Pay per token.
An always-on agent is a high-volume workload. Serverless inference bills per million tokens at the model library rates — no seat licenses, no minimums, and the calculator shows the monthly number before you commit.
Your agent reads your email. Zero data retention matters.
Agents handle the private stuff — inboxes, files, messages. Run BiOS processes prompts and responses in memory and never stores them, so the agent can work on sensitive material without it becoming training data for anyone.
Your weights, your endpoint, the same three lines.
Fine-tune a model on dedicated GPUs and serve it on a dedicated endpoint behind the same OpenAI-compatible API. The agent config does not change — only the model id does.
Questions teams ask before switching
Which agent frameworks work with Run BiOS?+
Any framework that accepts an OpenAI-compatible endpoint: Hermes Agent, OpenClaw, Cline, Aider, Continue, LibreChat, and the OpenAI SDKs themselves. The integration surface is three values — base URL, API key, model id — so if a framework can point at OpenAI, it can point at Run BiOS.
Do I need to change my agent’s code?+
No. You change where the framework looks, not what it does: set the base URL to the Run BiOS endpoint, use your API key, and pick a served model id. The framework keeps its own loop, tools, and memory.
Can my agent use a model I fine-tuned on Run BiOS?+
Yes. A finished checkpoint can go onto a dedicated endpoint behind the same OpenAI-compatible API. The agent config is identical — only the model id changes, and the weights behind it are yours.
How is agent traffic billed?+
Per million tokens, at the same input and output rates published in the model library. There are no per-seat or per-agent fees, and cached input is cheaper where the model offers it.
What happens to the data my agent sends?+
Prompts and responses are processed in memory and never stored. Run BiOS does not retain your data or train on it — the same guarantee whether the agent runs on serverless inference or on your dedicated deployment.
Point your agent at Run BiOS tonight.
New accounts start with a $10 free credit — enough to run an agent for days and see the three-value setup for yourself.
Per-million-token billing · Zero data retention · Your weights stay yours