AGENT LOOP

FROM AI CHAT TO OPERATING LOOP

Turn AI conversations into loops that hold up.

Build a loop where you already chat. Or distill one Claude conversation into a reviewed file, upload it to ChatGPT, and continue with the useful context intact. Agent Loop finds gaps, adds structure, and keeps you in control of what crosses platforms.

Improve a loop Read the source
  • V0.5
  • Open source
  • Role-agnostic
  • No data retention
INPUT / OUTPUT PUBLIC DEMO
01
Your loop or goalChatGPT · Claude · any MCP host
↓
02
Gap checkQuestions · missing steps · assumptions
↓
03
Operational loopRoles · handoffs · controls · exits
↓
04
Portable artifactValidated · deterministic · exportable

same normalized input → same loop identity

01 / THE POINT

A useful loop should outlive the chat where it was invented.

Agent Loop turns conversational intent into an explicit artifact that a person or compatible agent can inspect, improve, and reuse. If you do not have a loop yet, it recommends a starting structure.

02 / DIRECT VALUE

Three simple ways in.

V0.5 adds one-conversation Claude to ChatGPT handoffs. Public plugin directory review is still ahead.

BRING A LOOP

Improve what you already designed.

Continue in ChatGPT, Claude, or another agent and ask Agent Loop to find missing steps, handoffs, controls, failure paths, and stopping conditions.

Today: The host passes reviewed current steps through MCP. Agent Loop matches what you keep and shows what it adds.

START FROM A GOAL

Find the loop you should have considered.

Describe the outcome, task, and relevant risk. Agent Loop reflects the context, asks at most two questions, and recommends a structured starting loop.

Today: Recommendations come from a finite, versioned pattern catalog. This is not an unlimited AI-generated playbook.

CLAUDE TO CHATGPT

Carry the useful context, not the transcript.

Claude creates a reviewed .agentloop.json file. Upload it to a ChatGPT chat, inspect it, and choose what to use.

Today: One conversation at a time. No account export, automatic sync, history import, or memory write.

USE YOUR TOOL

Keep working where you already think.

Connect from an MCP-capable agent host. The loop artifact is structured and portable instead of being trapped in one conversation or vendor interface.

Today: The remote MCP server and plugin package are ready for testing. The public directory listing is not yet live.

ANY ROLE

Built around the work, not the title.

Use it for personal routines, team delivery, research, creative work, decisions, incidents, or controlled operations. No executive role is privileged.

Today: Everyday roles work directly. If no role is given, the accountable person defaults to you.

03 / HARD LIMITS

The demo is deliberately small.

Operationalize has a precise meaning in V0.5: create a usable operating artifact or reviewed handoff. It does not mean Agent Loop runs the work or synchronizes accounts.

WHAT IT DOES NOW

  • Starts from a loop or a goal. Improve an existing design or build a structured first version.
  • Works across roles. The task and context matter; the product is not a CTO-specific copilot.
  • Checks what is missing. Current steps are matched against the right pattern. Missing checks, handoffs, follow-ups, and exits become explicit.
  • Produces one portable artifact. Same normalized request, same loop identity.
  • Transfers one reviewed conversation. Claude distills it into a file that ChatGPT can inspect in the current chat.
  • Connects through remote MCP. The shared free endpoint may sleep and is rate-limited.

WHAT IT DOES NOT DO

  • It does not execute the compiled loop or operate your systems.
  • It does not read ChatGPT, Claude, or another platform's memory by itself.
  • It does not import chat history, synchronize accounts, or write platform memory.
  • It does not save loops, transcripts, or user profiles.
  • It does not authenticate authority or grant entitlements.
  • It is not yet listed in the public plugin directory and does not claim production security controls.

04 / HOW IT WORKS

From conversation to operation.

  1. 1

    Choose a starting point

    Describe a goal, show a reviewed routine, or ask Claude to create a handoff from the current conversation. Raw transcripts and sensitive data stay out.

  2. 2

    Review before anything moves

    Confirm Claude's distillation before download. In ChatGPT, inspect the uploaded file and resolve any differences before using it.

  3. 3

    Use it in the current chat

    Continue with the reviewed context, design a candidate loop, or manually copy selected memory candidates. Nothing is executed or remembered automatically.

POLICY›AUTHORITY›RISK›TASK›DEFAULTS

05 / PRICING DIRECTION

No checkout. No fake urgency.

These are pricing hypotheses for the product we intend to build. Only the open-source compiler and public demo exist today.

AVAILABLE NOW

Open source

$0

Self-host it. Inspect everything.

  • Deterministic compiler
  • Current pattern catalog
  • Local and remote MCP server
  • Public shared demo
  • No hosted SLA or support
Get the code

PLANNED / PRICE HYPOTHESIS

Team

$49/workspace/mo

For teams that need durable, private loop improvement.

  • Private hosted workspace
  • Version history and comparison
  • Custom pattern catalogs
  • Review, improve, and export
  • Basic support
Not for sale yet

FUTURE

Controlled

Custom

For regulated deployments with real identity and policy controls.

  • Approved identity provider
  • Private networking
  • Trusted capability registry
  • Central audit integration
  • SLA and deployment support
Design partner stage

06 / TRY IT

Keep your agent.
Add the loop layer.

The anonymous endpoint is for synthetic demonstrations only. Do not send personal, customer, credential, or production policy data.

REMOTE MCP SERVER

https://getagentloop.ai/connect/mcp
  1. Add the URL as a remote MCP app or custom connector in Claude or another compatible client.
  2. In Claude, ask for a reviewed handoff, confirm the summary, and download the .agentloop.json file.
  3. Upload the file into ChatGPT and say: “Inspect this Agent Loop handoff and help me continue here.”
CLAUDE STARTER PROMPT

“Create a reviewed Agent Loop handoff from this conversation for ChatGPT. Show me exactly what will be included before creating the file.”

ChatGPT setup guide Check service health