Hosting for AI applications and agents
AI applications rely on long-running requests, background workers, queues and vector search. 9live runs these as persistent processes on dedicated CPU, with PostgreSQL and pgvector and Redis on the same private network.
Common problems
- Agent runs and LLM tool chains can take minutes, longer than serverless function limits allow.
- Cold starts reload models, tokenizers and SDK clients on every invocation.
- The application, vector database and queue each live with a different provider, adding latency and separate bills.
How 9live helps
- Persistent workers
- Run agents and job processors as long-lived processes with no execution time limit.
- pgvector
- Store embeddings next to relational data in managed PostgreSQL, under the same backups.
- Queues on Redis
- BullMQ, Celery, RQ or arq against managed Redis on the private network.
- Memory headroom
- 32 or 48 GB of reserved memory keeps caches, embeddings and smaller models in memory.
- Streaming responses
- Server-sent events and WebSockets stay open for as long as the response takes.
Suggested plan
Grow's 12 vCPU and 32 GB, with PostgreSQL and Redis, fit most products built on hosted model APIs. GPU inference needs a Custom plan or a GPU provider.
12 vCPU · 32 GB · 320 GB NVMe · 2 managed databases
Questions
Do you offer GPUs?
Only on Custom plans, in limited quantities. Most teams call a model API or a GPU provider for inference and run the application, database and queues on 9live.
Can we run open-weight models on CPU?
Small models run acceptably on dedicated cores for internal tools and batch jobs. For production inference at volume, use GPUs.
Talk to an engineer about your setup
Tell us what you run today and where it runs. We reply within one business day with a recommended plan and an estimate for the migration.