Badgr — GPU Compute for AI Agents Copy
OfficialServer Configuration
Describes the environment variables required to run the server.
| Name | Required | Description | Default |
|---|---|---|---|
| BADGR_API_KEY | Yes | Your Badgr API key from https://aibadgr.com | |
| BADGR_BASE_URL | No | API endpoint override | https://aibadgr.com/v1 |
| BADGR_MAX_PRICE | No | Hard $/hr cap applied to all jobs | 5.0 |
| BADGR_MAX_RUNTIME | No | Max minutes for badgr_serve, deploy_model and launch_agent_task | 60 |
| BADGR_MAX_JOB_COST | No | Hard USD cap on one launch_agent_task | 5.0 |
Instructions
Guidance the server publishes about itself, which clients place ahead of the tool catalog so the model reads it before choosing anything.
This server publishes no instructions, or was last inspected before Glama recorded them.
Capabilities
Features and capabilities supported by this server
Protocol revision2025-11-25
| Capability | Details |
|---|---|
| tools | {} |
Tools
Functions exposed to the LLM to take actions
| Name | Description |
|---|---|
| badgr_runA | Run a one-off GPU job via Badgr. Returns deployment_id, status, receipt_id, logs_url. Hard limits: max_price_per_hour ≤ $5/hr. Always confirm with the user before running jobs that cost > $1/hr. |
| badgr_serveA | Provision a persistent OpenAI-compatible endpoint for an LLM model. Returns endpoint_url, deployment_id, receipt_id. The endpoint stays up (billing runs) until badgr_down is called. Hard limits: max_price_per_hour ≤ $5/hr. max_runtime_minutes is REQUIRED — agents must not create open-ended billing. Always confirm with the user before serving. |
| badgr_statusA | List active Badgr GPU deployments for this account. |
| badgr_logsC | Fetch logs for a Badgr deployment. |
| badgr_downB | Terminate a Badgr deployment and stop billing. |
| run_workloadA | Run a one-off GPU job via Badgr. Returns deployment_id, status, receipt_id, logs_url. Hard limits: max_price_per_hour ≤ $5/hr. Always confirm with the user before running jobs that cost > $1/hr. |
| deploy_modelA | Provision a persistent OpenAI-compatible endpoint for an LLM model. Returns endpoint_url, deployment_id, receipt_id. The endpoint stays up (billing runs) until badgr_down is called. Hard limits: max_price_per_hour ≤ $5/hr. max_runtime_minutes is REQUIRED — agents must not create open-ended billing. Always confirm with the user before serving. |
| check_resultA | Check the state of one Badgr deployment and fetch its most recent log lines. Use after run_workload or deploy_model, with the returned deployment_id. Call badgr_down when the deployment is finished so billing stops. |
| ai_chatB | Call the Badgr AI API (OpenAI-compatible chat completion) and return the reply with token usage. max_tokens is capped at 4096 per call. Usage is metered. |
| launch_agent_taskA | Launch an autonomous coding agent on a GitHub repository. The job is verified by running |
| check_launch_taskB | Check an agent task started with launch_agent_task: status, cost so far, verification output, receipt and recent log lines. |
Prompts
Interactive templates invoked by user choice
| Name | Description |
|---|---|
No prompts | |
Resources
Contextual data attached and managed by the client
| Name | Description |
|---|---|
No resources | |
TDQS
Scored across 11 tools
The set contains exact duplicate purposes: badgr_run and run_workload both run a one-off GPU job with identical descriptions, and badgr_serve and deploy_model both provision a persistent endpoint identically. badgr_status and check_result also overlap on deployment state inspection, making misselection highly likely.
Naming mixes a badgr_ prefix (badgr_run, badgr_status, badgr_logs) with unrelated generic verb_noun names (run_workload, deploy_model, check_result) and an unaffixed ai_chat. The duplication across naming schemes makes the surface feel inconsistent rather than deliberately organized.
Eleven tools is a reasonable raw count for the GPU compute domain, but two pairs are exact duplicates, so the effective surface is smaller and the count overstates the real capability. The set would be better scoped if duplicates were removed.
The lifecycle is broadly covered: run jobs, serve endpoints, check status/logs, tear down billing, call chat, and launch/check autonomous agent tasks. Minor gaps exist around pricing/model discovery or richer cost inspection, but core workflows are reachable.