Cerver
Infrastructure for AI sessions. One API for transcript, model, and compute.
The story
Overview
Cerver provides infrastructure for managing AI agent sessions at scale. It abstracts away the complexity of running agents by treating each session as a unified object containing transcript, model, compute, and cost information. Users can start with hosted sessions in seconds or attach their own machines to leverage existing subscriptions. The platform enables intelligent model routing through customizable policies, allowing teams to run cheaper models for routine work and reserve expensive frontier models for complex tasks. Key capabilities include mid-session model and compute swaps while preserving transcript history, side-by-side agent comparisons, and spending controls with monthly caps. Cerver is designed for developers building AI-powered applications, teams managing multiple agents, and organizations looking to optimize AI infrastructure costs.
Key features
Model Routing Policies
Automatically or manually route tasks to the right model based on complexity, saving up to 62% on compute costs by avoiding unnecessary frontier model usage.
Mid-Session Swaps
Switch models and compute providers during an active session while preserving the transcript, tools, and session identity.
Agent Comparison
Run the same task on two different agents in parallel within a single session to objectively compare performance.
Hybrid Execution
Start sessions in the cloud instantly, then attach your own machines when needed to use existing Claude Max or ChatGPT subscriptions.
Session Visibility
Monitor thousands of parallel agents in real-time with per-session dashboards showing chat, model, compute, and cost metrics.
Spending Controls
Monthly spending caps enabled by default prevent runaway costs, with metered billing at $2 per 1M tokens and zero charges for unused capacity.
Use cases
- 1
Developers
Ship AI features faster with a session backend that handles transcript management, model selection, and compute orchestration.
- 2
AI Teams
Give every team member real model flexibility to choose the right agent and compute for their work without cost surprises.
- 3
FinOps Teams
Optimize AI spending by routing routine work to cheaper models and tracking per-session costs across thousands of agents.
- 4
Organizations at Scale
Run thousands of parallel AI agents with unified visibility, billing, and spending controls across multiple teams and applications.
FAQ
How much does Cerver cost?
Cerver offers a $5 free tier for getting started. Hosted sessions are metered at $2 per 1M tokens billed monthly. Running agents on your own machines with existing Claude Max or ChatGPT subscriptions has marginal cost of zero.
Can I switch models mid-session?
Yes. You can swap the model and compute provider during an active session while keeping the transcript, tools, and session identity intact.
What models and compute providers does Cerver support?
Supported models include Claude Opus, GPT-5, Grok, and Gemma. Compute providers include Vercel, E2B, Cloudflare, Modal, and your own machines.
How do I get started?
You can start online in seconds with no setup or card required, or install locally with one command to connect your machine and use existing subscriptions.
Does Cerver have spending limits?
Yes. Every account ships with a monthly spending cap enabled by default that stops runaway agents, preventing unexpected bills.
Tech stack & tags
Feedback & Discussion
Discussion
Sign in to leave a comment or rating.
No comments yet. Be the first.