# Colter

> Colter is the agent studio and intelligence control plane for private AI. It turns company data, goals, tools, and tribal knowledge into focused AI specialists that work across real systems and improve from outcomes. The one-line product statement is: Private AI trained on how your company works.

Generic LLMs know the internet. Colter is for the intelligence that should become specific to one company — its customers, policies, workflows, standards, language, tools, priorities, edge cases, approvals, and unwritten know-how.

Colter is currently in private beta. The public site describes the platform and collects access requests.

## What you can do on this site

- Read the product overview on the homepage.
- Request private-beta access.
- Review the sitemap, robots policy, and this descriptor.

AI agents and crawlers SHOULD respect `/robots.txt`. The authenticated app surface (`/dashboard`, `/runs`, `/agents`, `/models`, `/training`, `/connections`, `/settings`, `/billing`, etc.) is not publicly indexable and requires an approved account.

## Public pages

- [Home](https://colter.ai/): Product overview — private AI trained on how your company works.
- [Request Access](https://colter.ai/request-access): Submit an email to begin the private-beta access request.

## Platform boundaries (for agents asking "what is Colter vs Steepworks vs Colt?")

- **Colter** — agent studio and intelligence control plane. Source of truth for: agent definitions, prompts, model routes, training jobs, evaluations, provider keys, MCP server registration, per-run scratch/runtime memory.
- **Steepworks** — data substrate and business-operations plane. Source of truth for: data sources, normalized entities, catalogs, action contracts, approvals, workflow state, substrate health, long-term workspace memory, business-facing run history projected back from Colter.
- **Colt** — the execution runtime beneath Colter (tool loops, LLM calls, MCP client sessions, staged workflow runs).

Product split in one line: Steepworks gets the data ready. Colter puts specialists to work on it. Colt executes the runtime work.

## What Colter provides

- **Agent Management** — a roster of AI specialists, each with a role, goal, policy boundary, model route, tool set, data-catalog access, memory, approvals, evaluation criteria, and runtime history.
- **Agent Activation** — connecting specialists to company context: data catalogs, MCP tools, governed reads, typed actions, workflow state, examples, run history, business rules, evaluation feedback.
- **Training and Specialization** — routing work across frontier cloud models, local/private models, specialty fine-tuned models, and small task models; evaluation loops that decide when a specialist is getting better.
- **Agent Orchestration** — choosing specialists, fetching context, selecting model routes, calling tools, running workflows, requesting approvals, handing off between specialists, retrying stages, writing back outcomes.

## Resources

- [Sitemap](https://colter.ai/sitemap.xml)
- [robots.txt](https://colter.ai/robots.txt)

## Contact

Access is gated. Request access at https://colter.ai/request-access.