Your GPU capacity
Model and inference stay on your hardware
How does one request travel?
Your app
model: "byoc/office-llm"
BazaarLink BYOC
Forwards to your node
① Your node
Your GPU node
Model fee: zero
② Fallback (per your setting)
BazaarLink platform
Billed to: Platform credits
Your app sends the request as usual. No code change.
Illustrative animation. The agent connects outbound to the gateway, so no inbound port is needed. First bind byoc/<slug> to your local model ID under Model bindings.
Customer machine → BYOC Agent → BazaarLink gateway → Customer app
The model and inference environment stay on your equipment. The Agent opens the gateway connection, the gateway routes your account’s work to your node, and the app keeps one API entry point.
BazaarLink models
Can take over when your node is full or offline, per your fallback settings (platform rates)
BazaarLink BYOC
No BazaarLink model charge for traffic your own node handles
One API key
Same endpoint, no client changes
Your application
Local backends supported today
Choose the local backend that runs your model, then pass its connection settings to the Agent.
Ollama
Default backend; the default URL is http://localhost:11434.
LM Studio / vLLM
Use --backend openai and --url with the model ID your local server accepts.
llama.cpp server
Uses the same OpenAI-compatible streaming path without model-name translation.
Why connect compute to your own account
Models run on your equipment
You control the model files, inference process, hardware, and local runtime.
No BazaarLink model charge for this traffic
For traffic your own BYOC node handles, BazaarLink doesn't charge a model fee; if you also use a platform feature such as web search, that feature is billed at platform prices. Equipment, power and operations remain yours.
One API for the application
The local model connects through the gateway, so product code does not need a transport layer per backend. The node currently receives requests for /v1/chat/completions, /v1/responses, /v1/messages, and image generation (/v1/images).
The node serves its own account or organization
Work goes only to nodes bound to that account or organization. The node is not placed in a shared pool or compute marketplace.
Your node handles more than chat
Image generation
The Agent forwards image generation requests to your backend's /images/generations.
Tool calling
tools and tool_choice are forwarded to your backend, and tool_calls come back in OpenAI format (OpenAI-compatible and Ollama backends).
Reasoning content
Reasoning models return their thinking in the reasoning_content field.
Scope and fallback
On the Fallback Settings page, choose which BazaarLink API keys this node applies to. When the node is full, offline or errors, the request tries the next source in your fallback order; requests handled by a platform model are billed at platform rates.
Web search works on your own compute
Requests your node handles can add :online too, so the model checks the web before answering: the search runs on the platform side and the results go to your node together with the question. The model fee is zero; only the search fee is charged, at platform prices, so your account needs enough balance to cover the search.
Entry protection
You manage the local logs for BYOC; before a request reaches your GPU, the platform entry still applies moderation, and sensitive-data filtering is available as an opt-in.
Platform moderation
Platform moderation runs at the API entry (/v1/chat/completions, /v1/responses) before the request is processed; BYOC cannot disable it. Blocked requests do not reach your node.
Sensitive-data filtering you can enable
This is opt-in: organization keys use the organization rule, while personal keys are enabled by the key owner at /content-filter. Once enabled, inbound prompts replace API keys (OpenAI/Anthropic/GitHub/Google/AWS), JWTs, private keys, credit-card numbers, and Taiwan national ID numbers with [REDACTED:TYPE] before the node; prompt injection is blocked (403, code: content_filter).
Enable at /content-filterFrom GitHub install to connected node
@bazaarlink/byoc-agent v0.2.0 is MIT licensed. Install the CLI globally from the public GitHub repository; the npm registry version is coming later. Authenticated only confirms the gateway connection. Then open /keys/byok?tab=byoc, map a canonical API ID such as byoc/<lowercase-slug> to the exact model ID accepted by your local backend, and call that canonical ID through the API.
01
Install the CLI globally
Install the MIT-licensed CLI directly from the public GitHub repository.
npm install -g github:Bazaarlinkorg/bazaarlink-byoc-agent02
Get a BYOC key
If you already have one, continue; otherwise use register with your account email.
bazaarlink-byoc register \
--email you@example.com \
--max-concurrent 403
Log in to the gateway
login calls /auth/byoc/login and stores the key plus a short-lived token locally.
bazaarlink-byoc login \
--key "byoc_..." \
--gateway "https://byoc-gateway.bazaarlink.ai" \
--input-price 0 \
--output-price 004
Start and connect
Authenticated only confirms the gateway connection. Then open /keys/byok?tab=byoc, map a canonical API ID such as byoc/<lowercase-slug> to the exact model ID accepted by your local backend, and call that canonical ID through the API.
bazaarlink-byoc start
# Authenticated only confirms the gateway connection.
# In /keys/byok?tab=byoc, map byoc/<lowercase-slug> to exact backend ID.Technical details
These fields mirror the current CLI and package metadata. Install from the public GitHub repository until the npm registry version arrives.
- npm package
- @bazaarlink/byoc-agent v0.2.0 · MIT · install from GitHub, npm registry version coming later
- Source
- github.com/Bazaarlinkorg/bazaarlink-byoc-agent
- CLI
- bazaarlink-byoc
- Runtime
- Node.js 18+
- Connection
- Agent opens gateway /ws outbound; no inbound port is required
- Local config
- ~/.bazaarlink/config.json (Windows: %USERPROFILE%\.bazaarlink\config.json)
- login pricing flags
- --input-price and --output-price are currently required; 0 is a compatibility value, not a BYOC charge
The CLI is installed globally from the public GitHub repository; the npm registry version is coming later.
--backend ollama is the default. Add --ollama-url when Ollama runs elsewhere.
LM Studio, vLLM, and llama.cpp use --backend openai; --url is required and After connecting, map a canonical API ID to the model ID accepted by your backend in the BYOC bindings page.
Who this is for
You have a GPU host and local model, and want the model files and inference environment to stay on equipment you control while your product keeps a familiar API entry point.
You do not want to rent out idle compute or place requests into a shared node pool. BYOC connects your equipment, your account, and one gateway path.
Connect your first BYOC node
Bring one local model machine and a BYOC key, then validate how your compute, your node, and one API entry point work together.
Manage BYOC nodes