The operations layer and workflow engine for AI agents
Describe the process once.
Your agents run it, step by step.
Claude Code, Codex, Cursor, ChatGPT or any MCP client runs your process one step at a time. Gates they cannot skip, one shared memory, every change audited.
At a glance
The operations layer and workflow engine for AI agents
- Status
- Available now. Free plan, no card.
- Built for
- Engineering leads, Heads of AI, operations owners and teams running AI agents on real work
Why it matters
Positioning pillars
Workflow Engine
A step is not a prompt.
Your process becomes a workflow your agent walks one step at a time. Gates are evaluated by the engine, decisions can invoke sub-workflows that return, fix loops go back to an earlier step, and parallel work runs as child tasks. One router workflow sends every new task to the right process.
Brain
Run 40 beats run 1.
Memories hold what the team learned. Routes hold where things live. Agents recall before they act and capture on the way out, so the next run starts with what the last one found. One recall spans every workspace.
Governance
The run stops. You decide.
Approval gates hold the run until a person signs off, and no prompt skips them. Policies reach the agent at the moment it acts. Every change writes an audit row that keeps the agent's own label apart from the server-verified human.
Autonomy
Attended today. Unattended when you say so.
Start attended in the AI session you already use, with no backend to install. When a workflow is proven, schedule it and let executors pull the work: on your machine, on your own runner, or one isolated pod per run on your Kubernetes.
Bring Your Own AI
ConvOps runs no models. You keep yours.
ConvOps is an MCP server, so it connects to Claude Code, Claude, Codex, ChatGPT, Cursor, OpenCode or any MCP client with one line of config. Swap the client and the process stays. You own the execution layer. ConvOps holds the process and the record.
The mechanics
How it works
Step 1
Connect your AI client
Add the ConvOps MCP server to the client you already use. One line of config, a browser sign-in on the first call, nothing to migrate.
Step 2
Describe the process in conversation
Tell your AI how the work gets done. It authors the workflow: steps, gates, decisions, loops. The engine validates the workflow before it ever runs.
Step 3
Agents run it one step at a time
The agent receives only the current step, never the whole process. Gates hold the run until their condition is met. Work resumes in any session, with any client, exactly where it stopped.
Step 4
Every change lands in the audit trail
Each step, decision and approval is recorded with who made it and what changed. When someone asks how a piece of work was done, the answer already exists.
Neomanex internal operations
Neomanex runs its own engineering, content, sales and operations on ConvOps every day: every task, approval and release goes through a workflow. The product is the system we built to run ourselves.
The contrast
How we compare
Them
One giant system prompt
Us
Small workflows the agent walks one step at a time. Graph engineering, not prompt engineering.
Them
An agent framework you code against
Us
No code. The process is a workflow any MCP client can run.
Them
Cron jobs that fire and forget
Us
A governed loop: schedules, pools and gates, with every run recorded.
Them
Project management with AI bolted on
Us
The agent does the work inside the workflow. The board is the side effect.
Them
An AI platform that locks you to one model
Us
ConvOps runs no models. Bring the AI clients your team already uses.
The ecosystem
Fits into the portfolio
Related products
Questions, answered