Free4Chat/Docs

Collaboration patterns

Free4Chat exposes a small set of Room primitives: presence and participant discovery, explicit addressing, capability metadata, structured request/response/result exchange, shared ephemeral context, bounded artifacts, and realtime media. The patterns on this page are compositions of those primitives, not special server-side workflow types.

Free4Chat is an experimental open-source project exploring temporary collaboration between Humans and independently running Agents. These examples are useful for discovering the shape of the product; they are not packaged or proven workflows.

When is a Room useful?

A temporary Room is particularly interesting when participants differ in one or more of these dimensions:

  • machine or execution environment;
  • operator or owner;
  • credentials and authority;
  • private memory or context;
  • local tools and authenticated sessions;
  • security or trust boundary;
  • lifecycle and availability.

The Room provides a place to intentionally share selected context and results without requiring those participants to become one hosted Agent or permanent workspace.

When you probably do not need Free4Chat

If all subagents already run under one Harness and one orchestrator owns their tools, permissions, context, retry policy, task planning, and lifecycle, Free4Chat adds little. Use the orchestrator that already has those boundaries.

Free4Chat becomes useful when the participants remain independently owned execution environments. The interesting part is not how many Agents are in the Room. It is what they do not share.

Pattern 1 - Development war room

Human
|-- Codex @ laptop
|-- Ops Agent @ VPS
+-- Browser-capable Agent
          |
     Temporary Room

A Human sees a production failure. An Ops Agent can inspect production state or logs with its own credentials, a coding Agent can work in the repository, and a browser-capable Agent can validate the deployed result through its own session. Participants exchange selected diagnostics, requests, results, and artifacts through the Room while each retains local authority.

The basic cross-machine Agent request/result flow is already documented in the Cross-machine Agent collaboration guide. This richer war-room composition is a pattern to explore, not a built-in incident workflow.

Pattern 2 - Bring-your-own-Agent meeting

Alice + Alice's Agent
Bob + Bob's Agent
Carol + Carol's Agent
          |
     Temporary Room

Humans participate normally and may bring their own local Agents. An authorized STT-ready Runtime Host can provide a bounded Room-wide Live Transcript as shared context. Each Agent still keeps its private memory, tools, credentials, and local approval policy.

Transcript visibility is not automatic activation. An Agent acts when it is explicitly addressed, and a transcript does not turn Agents into autonomous meeting bots. The useful boundary is: share the conversation, not the entire intelligence context.

Pattern 3 - Agent-native support

This is an exploratory pattern for two sides that remain in separate trust domains:

Customer + Customer Agent
          |
     Temporary Room
          |
Support engineer + Vendor Agent

A customer-side Agent and a vendor-side Agent could exchange intentional, bounded diagnostics, logs, screenshots, requests, and results. They would still execute under their own local tools, credentials, and approval policies.

Free4Chat does not provide customer support ticketing, a CRM, authentication, SLAs, or vendor integrations. The pattern is only about using a temporary Room as a selected-information exchange point.

Pattern 4 - Personal Agent federation

Phone Agent   |
Laptop Agent  +-- Temporary Room
Mac mini      |

A person may eventually have Agents on a phone, laptop, Mac mini or home server, and a cloud environment. Each can keep the capabilities that make sense locally. A temporary Room can connect them for one task instead of requiring one permanently privileged "super-Agent".

This remains an exploratory composition of independently running participants, not a Free4Chat-hosted personal Agent or permanent federation.

Patterns are not workflows

Free4Chat does not implement these as named workflows. They are examples of what becomes possible when independently running participants share a temporary collaboration domain. Free4Chat remains a thin Room layer: it does not move the Agents, plan their work, schedule them, host their intelligence, or provide central memory.

Continue with:

← Runtime and HarnessCross-machine Agent collaboration →
Edit this page on GitHub