> ## Documentation Index
> Fetch the complete documentation index at: https://docs.formation.texturehq.com/llms.txt
> Use this file to discover all available pages before exploring further.

# What is Formation?

> Agents that participate in shared conversations, remember context, and follow through on work.

Formation runs agents that work with people in shared conversations. An agent has its
own name, profile, accounts, tools, and memory. It can participate in Slack,
Telegram, or phone messaging while keeping track of who is speaking and which
conversation a request belongs to.

One Formation installation can host several agents. Each keeps a separate perspective:
hosting two agents together does not give them access to one another's private
conversations, memories, or credentials.

## Start with one agent

If you already have access to a Formation installation, follow
[Get started](/quickstart.md) to create an agent, connect it to a conversation, and
check its first reply. If you are running the installation, start with
[Installation](/operations/installation.md).

For example, a team can give an agent its own Slack identity, connect a project
tool, and ask it to research an issue. Formation records the messages, model calls,
tool results, and reply so an operator can inspect what happened. Additional
capabilities depend on the providers and permissions configured for that agent.

## What you can do

| Need | Formation capability |
| - | - |
| Keep context about people and past work | [Memory](/runtime/memory-retrieval.md), [actor profiles](/runtime/actor-profiles.md), and [conversation search](/runtime/conversation-search.md) |
| Learn from recent conversations in the background | [Dreaming](/runtime/dreaming.md), with opt-in scheduling, review, and undo |
| Track work that outlives one reply | [Commitments](/runtime/commitments.md), with follow-up, blockers, evidence, and closure |
| Send recurring reminders or briefings | [Schedules](/runtime/schedules.md), with occurrence history and operator controls |
| Choose a model for each kind of turn | [Automatic model selection](/runtime/adaptive-model-routing.md) and [decision models](/runtime/decision-models.md) |
| Use external services and files | [Tools and capabilities](/runtime/tools.md), [remote MCP](/runtime/remote-mcp.md), and [document reading](/runtime/document-reading.md) |
| Give an agent a service credential | [Vault](/runtime/agent-vault.md), with use by name in the browser or workspace |
| Investigate an unexpected response | [Inspect agent activity](/runtime/observability.md) |

These are distinct features. Memory stores reusable information; a commitment
tracks an obligation; a schedule starts work at a defined time; Dreaming reviews
past evidence. Enabling one does not enable the others.

## A few terms

| Term | Meaning |
| - | - |
| Agent | A persistent identity with its own profile, models, capabilities, and perspective. Code also calls it a `RuntimeAgent`. |
| Actor | A participant as known to one agent. Verified links can connect that person's accounts across platforms. |
| Conversation | A room, direct message, or thread captured through a connector. |
| Connector | An adapter for receiving and sending messages on a platform. |
| Turn or run | One bounded execution of an agent in response to a message or other admitted trigger. |
| Operator | A person who administers Formation through the dashboard or API. |

The [architecture overview](/layers.md) explains how these fit together. The
[design principles](/philosophy.md) explain why identity and explicit state matter.

## Which version do these docs describe?

The published guides describe the implementation on the repository's `main`
branch. An installation may run an older build or have different capabilities
enabled. Owners can check **Settings → About this install** for the build and
database state; builds without a commit stamp report that explicitly.

Engineering proposals remain in the repository and are labelled as such. In
particular, a general WorkGraph scheduler and organization membership model are
not available product features.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.