Workflows
Every job to be done — a chat, a heartbeat, a transaction, a process — runs as a workflow powered by agents and sub-agents.
Telos Brain
Everything the brain is made of.
Workflows, skills, memory and learning — plus the infrastructure to run it, control cost, and keep it secure.
What's inside
Every layer of the brain.
From the core building blocks to how you deploy, integrate and keep it secure.
01
The Brain Itself
What a brain is made of. The core building blocks.
02A Brain That Learns
The self-learning system. This is how you train your brain.
03Built for Real Organisations
Multi-tenancy, scale, and operational visibility.
04Cost Under Control
See what your AI costs — and cut it.
05Working with LLMs
The plumbing most teams don't know they need — handled.
06How You Build & Deploy
Local-first development, one-command cloud deploys.
07APIs & Integration
Everything talks to the brain through a secure API.
08Security, Trust & Licensing
Encryption, isolation, versioning, and a licence you can explain.
01
The brain itself.
What a brain is made of. The core building blocks.
Skills & Skill Books
Knowledge lives in portable, versioned skill books — accounting, project management, software delivery — that copy between brains in seconds.
Tools as Mini-Skills
Tool definitions are first-class citizens of the brain schema; skills can even introduce their own tools into a run.
Memory
Memory is tracked at the brain level, per entity, and as managed context on every unit of work.
Entities
Keep data and memory fully siloed between clients, departments, or teams — tenancy built into the brain's DNA.
Units of Work
Track any job or ticket as a long-running unit of work with its own managed context, data logs, and multiple workflow runs against it.
Brain DNA in One Folder
Your entire brain — workflows, skills, tools, memory, configuration — is defined as a single schema folder you own and version in source control.
02
A brain that learns.
The self-learning system. This is how you train your brain.
Learning Inbox
Every learning — from evals, users, emails, or uploaded expert opinion — lands in one inbox to be triaged.
Apply Learnings Anywhere
Triaged learnings update skills, tools, workflows, or memory — automatically, or through an approval process you control.
Built-in Evals
Eval workflows grade every run 0–100, with traffic-light grades and trends visible in the console.
Train by Email or Upload
Teach your brain by emailing it content or uploading expert knowledge — no technical work required.
Checkpoints, Diffs & Rollback
As the brain learns and changes itself, view diffs between any two checkpoints and roll back with confidence.
Task Assignment
When the brain needs a human, it can assign inbox tasks to named team members for review and approval.
Starter Brain
Pull down a working, self-learning brain that already knows how to build and manage itself — including a skill book on running the brain.
03
Built for real organisations.
Multi-tenancy, scale, and operational visibility.
Two-Level Multi-Tenancy
Entities silo data between clients or departments; units of work isolate context per job — both native to the platform.
Multiple Brain Instances
Deploy as many instances of your brain as you need — per environment, per model, per client.
Clone & Template Updates
Deep-clone a brain in one call, or push configuration updates from a template brain to a fleet of live brains.
Full Telemetry & Visualisations
Every run captures rich telemetry — LLM activity plus user events — with visualisations out of the box.
30-Day Data Retention
Run history is retained for 30 days so you can run evals and analyse brain performance over time.
04
Cost under control.
See what your AI costs — and cut it.
Cost Visibility Per Run
Understand exactly what each workflow run costs, and where the money is going.
Waste Detection
Spot cache misses, failure rates, and retry waste — then fix them through configuration, not code.
Proven Savings
Teams have cut LLM costs in half; the brain itself typically costs around 1% of your LLM spend.
Model-Optimised Instances
Switching models isn't just a dropdown — deploy a separate brain instance fine-tuned (tools, caching, compaction) for each model.
05
Working with LLMs.
The plumbing most teams don't know they need — handled.
Multi-Model Support
Claude, ChatGPT, Gemini, and Grok — plus local LLMs for fully on-premise development.
Automatic Caching
Caching modes managed automatically per provider, so you get the best cache hit rates from every model.
Model-Portable Compaction
Automatic context compaction at a configurable threshold, consistent across models — with a controllable compaction prompt, runnable as a tool.
Output-Token Retries
Automatic retry mechanism manages output token limits, so long generations don't silently fail.
Managed Embeddings
Embeddings across skills, tools, and memory are generated and maintained by the brain — vector search with zero setup.
Vision & Transcription
Extract text from uploaded files and images via a built-in transcription endpoint using your configured vision model.
Thinking Modes & Turn Control
Run telemetry captures model, thinking mode, and turn totals — with sensible failure handling on max-turns and token exhaustion.
06
How you build & deploy.
Local-first development, one-command cloud deploys.
Open-Source CLI
The command line interface is open source — init, develop, and deploy your brain from your terminal.
Open-Source Starter Brain
The starter brain schema repository is open source — fork it and go.
Run Locally in Docker
A Docker container ships the full brain infrastructure to run locally — connect it to a local LLM and develop entirely offline.
One-Command Cloud Deploy
Develop locally, then deploy to the Telos cloud with a single command.
Configuration as Code
The whole brain is YAML and markdown in a folder — versioned, diffable, reviewable like any codebase.
Self-Editing Schema
In-brain system tools let a running brain inspect, edit, and create its own schema — with brain scoping, exact-match edits, and automatic versioning as guardrails.
Works with Your Stack
Any environment with a sandbox can run the CLI — Lovable, Vercel, AWS, Azure — bring the starter brain and build your tool layer.
Powerful Variables System
Template tags scope variables across inputs, entities, and units of work — with input-tools that pre-call tools automatically at run start.
Interactive Onboarding
A getting-started interview configures your brain schema from scratch — entity type, unit of work type, categories — before first deploy.
07
APIs & integration.
Everything talks to the brain through a secure API.
Secure Execution API
All interaction with the brain runs through an authenticated API with consistent conventions, error formats, and tenancy resolution.
Sync & Async Runs
Run workflows synchronously with SSE streaming, or fire-and-forget asynchronously with callbacks.
Multi-Turn Chat Sessions
Hold a full multi-turn conversation against a single workflow run.
Management API
Clone brains, update from templates, and manage instances programmatically.
Connectors
Connect to external services via API key, OAuth 2, or MCP — declared in simple YAML with environment-specific base URLs.
System Tools Out of the Box
Skill search, list and retrieval; memory search; tool search; inbox operations; unit-of-work context and data; run grading; user lookup and task assignment.
Progressive Disclosure
Skills reference each other by code, so agents load exactly the knowledge they need, when they need it — managed automatically by the brain.
OTEL Telemetry
Full OpenTelemetry-format telemetry across all runs — plug into your existing observability stack.
08
Security, trust & licensing.
Encryption, isolation, versioning, and a licence you can explain.
Encrypted Secrets Store
Environment variables, API keys, and connector credentials are uploaded, encrypted, and injected into outbound requests — never in your schema.
Tenancy Isolation
Brain scoping is enforced at the tool and API level — every operation resolves to exactly one brain, one organisation.
SSRF Protection
Async callbacks are restricted to allow-listed domains.
Automatic Versioning
Every schema change is versioned automatically — nothing changes without a trail.
Clear Licensing
Open-source CLI and starter brain; MIT-licensed Docker runtime container; commercial cloud at US$2 per 1,000 minutes with unlimited brain instances.
Comparison
Telos Brain vs Eve, Claude and AgentCore.
A side-by-side look at models, skills, memory, tenancy, learning and lock-in.
Open the comparison ↗Design
UX design for AI brains.
How to design the custom app and the company brain together — jobs, knowledge, and the moments a person must decide.
Open the design guide ↗Telos Brain
Build momentum.
With wisdom.
Turn what your business knows into infrastructure your AI can use.
Let's build your Company Brain.