Telos Brain
A self-learning company brain: agents, skills, memory and multi-tenancy defined as portable configuration, running locally or in the cloud.
Comparison
Telos Brain compared with Eve, Claude and AgentCore.
If you are choosing an agent stack, this is the difference that matters: do you get a company brain you own, or a framework you have to assemble?
What they are
What each product is.
Public preview and beta releases, as at August 2026.
Vercel Eve
An open-source, filesystem-first framework for building durable agents that deploy to Vercel.
Claude Managed Agents
Anthropic runs the harness and infrastructure. You define tasks, tools and guardrails. Claude-only.
AWS AgentCore
Modular AWS services — Runtime, Memory, Gateway, Identity, Observability — that you assemble around your own agent code.
Features
Feature comparison.
Models, where it runs, skills, memory, tenancy, learning, cost and lock-in.
| Feature | Telos Brain | Vercel Eve | Claude Managed Agents | AWS AgentCore |
|---|---|---|---|---|
| Model support | Claude, ChatGPT, Gemini, Grok and local LLMs | Any model via AI Gateway | Claude only | Model-agnostic on Bedrock and beyond |
| Defined as | Config-as-code: one schema folder — the brain’s DNA | Filesystem-first markdown and TypeScript tools | API calls and console; Anthropic owns the harness | Your own code in containers |
| Runs locally | Full brain in Docker, including local LLMs — then one command to cloud | Local dev, deploys to Vercel | Anthropic cloud. Self-hosted sandboxes optional | AWS cloud only |
| Skills | Skill books: portable, versioned, cross-referencing, embedding-searched | Markdown skills per agent | Claude Skills supported | No native skills concept |
| Memory | Per brain, per entity, per unit of work | Session state and files | Built-in cross-session memory | Managed short- and long-term memory service |
| Multi-tenancy | Native two-level: entities and units of work | Build it yourself | Build it yourself | Session isolation only |
| Self-learning | Learning inbox applied to skills, tools, workflows and memory — with checkpoints, diffs and rollback | Not available | Partial: self-evaluating outcomes and dreaming, research preview | Not available |
| Evals | Built-in eval workflows grade every run 0–100 | Evals built in | Self-evaluation loop, preview | Bring your own |
| LLM cost optimisation | Per-run cost visibility, automatic caching, cross-model compaction, output-token retries | Token usage reporting only | Harness manages context, Claude only | Metrics only |
| Platform lock-in | Low: same schema runs local, on your infra, or Telos cloud | Vercel. Other adapters on the way | Anthropic cloud | AWS |
| Cloud pricing | US$2 per 1,000 minutes, plus your LLM costs | Vercel usage-based | LLM tokens plus $0.08 per session-hour | Per-service consumption |
As at August 2026. Eve is in public preview. Claude Managed Agents is in public beta; multi-agent coordination and outcome-based features are research preview. AgentCore can call external models; the infrastructure itself is AWS-bound.
Telos Brain
What the others do not include.
These are the parts of a company brain you would otherwise have to build yourself.
Clients stay separate
Tenancy is built in. Each client, team or job keeps its own data and memory.
It learns from the work
Useful lessons go back into skills, tools and memory — with checkpoints, so you can review or roll back.
Your ways of working travel
Skill books are portable. Copy proven methods from one brain to another in seconds.
Any model, less waste
Use Claude, ChatGPT, Gemini, Grok or a local model. Caching and compaction keep token cost down.
Same brain, anywhere
Run it on your laptop, on your own infrastructure, or on Telos cloud. The definition does not change.
Telos Brain
Build it yourself.
Or talk to us.
A custom harness plus company brain — on your tools, or run by Telos.