Should you choose HumanLayer or ConvOps?
Choose HumanLayer when your team wants an IDE for AI coding sessions. Choose ConvOps when the agents you already use must follow a process with sign-off, memory and a record.
choose HumanLayer if
- Your team does AI coding and wants one IDE for sessions, plans, diffs and design reviews.
- You want an opinionated, built-in method such as its QRSPI workflow.
- Inline comments on design docs by humans and agents are how you want to review.
- You need SOC 2 Type II today and per-seat pricing suits you.
choose ConvOps if
- You want to define your own process, not adopt a built-in one.
- The work includes non-coding processes: content, sales, operations.
- People use different AI clients, including ChatGPT and Claude desktop.
- You were using the HumanLayer SDK for approvals and need gates that are maintained.
How do HumanLayer and ConvOps compare?
Side by side, as of October 2026. ConvOps is the operations layer for AI agents: an MCP server that holds the process your AI clients follow, with approval gates, a shared memory and an audit trail. Every HumanLayer cell links to the official source it comes from; ConvOps cells describe the product as it ships today.
| Capability | HumanLayer | ConvOps |
|---|---|---|
| What runs the steps | Coding agents you bring (its FAQ lists Claude Code, Codex, Copilot and Fireworks), in sessions on your laptop or on cloud daemons. [1] | Your AI client runs each step. ConvOps hands it one step at a time and runs no AI models itself. |
| Where the process lives | Inside HumanLayer: tasks group sessions, artifacts and worktrees, with built-in workflows such as RPI, PRD-oriented and QRSPI. [1] | A workflow in your ConvOps workspace, reached over MCP and changed by asking your AI client. |
| Human approval gates | Comment-driven review of design docs and plans before implementation. The original approval SDK is deprecated. [2] | Approval gates on any step, evaluated by the engine. The run waits at the gate, for minutes or days, until approval is passed. |
| Memory | Versioned artifacts per task (research, design, plan). Pro lists "Compounding Engineering: agents get smarter over time". [1] | A shared brain of memories and routes, recalled across sessions, AI clients and workspaces. |
| Audit trail | Full session visibility of every message, tool call and code change. Audit logs are on the Enterprise plan. [1] | An audit row for every change to tasks and workflow runs, keeping the agent's self-declared label apart from the server-verified human. |
| Attended or unattended | A local daemon runs many sessions in parallel; cloud daemons handle long-running work and keep going when your laptop closes. [1] | Attended in your AI session, or unattended through executors and RRULE schedules, including one pod per run on your own Kubernetes. |
| Self-hosting | On-prem and private VPC on Enterprise. [1] | Hosted at mcp.convops.app, or self-hosted on your Kubernetes with one Helm chart on Enterprise. |
| Pricing model | Starter free (up to 3 members, 200 sessions a month), Pro $100 per user per month, Enterprise custom. Bring your own model keys or subscriptions. [1] | Free $0, Pro $79/mo ($66/mo billed annually), Business $299/mo ($249/mo annually), Enterprise from $2,500/mo. Every feature on every plan; you pay for scale. Your AI client is billed separately. |
| Client support | Its own IDE and cloud, with the same UI on desktop, web and phone. [1] | Any MCP client. Verified: Claude Code, Claude (web, desktop and Cowork), OpenAI Codex, ChatGPT, Cursor and OpenCode. |
What is HumanLayer today?
HumanLayer describes itself as "the multiplayer control plane for your software factory": a coding-agent IDE and cloud that brings agent sessions, plan artifacts and code diffs together for a team.
Tasks group agent sessions, artifacts and worktrees. Built-in workflows guide the work, among them RPI, a PRD-oriented flow, Oneshot and QRSPI (questions, research, design, structure, plan, implement), and design docs get inline comments from people and agents before code is written. A local daemon runs sessions on your machine; cloud daemons take long-running work (HumanLayer).
Many people still know the name from its earlier open-source SDK for human approval of agent tool calls. The GitHub repository now reads: "the code here is pretty much all deprecated", and points to the rebuilt product (GitHub). Per HumanLayer's FAQ, its Research, Plan, Implement (RPI) framework is open source and the product is not, yet (HumanLayer).
What is the main difference between HumanLayer and ConvOps?
HumanLayer is a place you work: an IDE with its own workflows for coding agents. ConvOps is a layer under the AI clients you already use, holding your own process for any kind of work.
In HumanLayer, review happens on artifacts: you comment on a design doc and the agent picks the comments up. In ConvOps, review is a step: an approval gate the engine evaluates, so the run cannot move past it until approval is passed. Both put people in the loop; they put them in different places.
ConvOps gates sit on workflow steps, not on individual tool calls. For per-call approval inside a session, use your client's own permission system alongside ConvOps. See human-in-the-loop.
Where is HumanLayer stronger?
As a team workspace for AI coding. ConvOps has no IDE and no session viewer.
Design review on documents
Inline comments by people and agents on research, design and plan artifacts, fed straight back to the agent.
Session visibility
Drill into every thinking message, subagent tool call and code change of a session.
Worktrees and multi-repo
Tasks provision worktrees across repositories and launch the first session for you.
Ready-made method
Built-in workflows such as RPI and QRSPI, with intake from Linear or Jira.
Compliance paperwork
SOC 2 Type II, with ZDR and DPA agreements on enterprise tiers.
Where is ConvOps different?
Your process, any work, any MCP client, priced by scale rather than seats.
Your own workflows
Describe any process; the router sends new work to the right workflow.
Engine-evaluated gates
Approval, finished child tasks and a required note are checked on every advance; refusals name the unmet condition.
Beyond coding
Content, sales and operations run on the same engine with the same record.
Every client
Verified with Claude Code, Claude (web, desktop and Cowork), OpenAI Codex, ChatGPT, Cursor and OpenCode.
Shared brain and audit
Memories and routes across workspaces; an audit row per change on every plan.
Can you use HumanLayer and ConvOps together?
They overlap on coding work, so pick one to hold that process. ConvOps can cover the processes outside the IDE either way.
If your engineers live in HumanLayer, ConvOps can still run the release approvals, content, sales and operations work around them. If you would rather keep coding in Claude Code directly, a ConvOps workflow can carry a QRSPI-style process with a gate at design review and another before ship. We have not tested ConvOps inside HumanLayer sessions.
one ConvOps workflow, with HumanLayer inside one step
How do you replace the HumanLayer approval SDK?
Move the approval from the tool call to the workflow step. Connect a client, describe the process and mark the steps that need a person.
- 1
Connect your AI client
In Claude Code, one line. OAuth sign-in on first use.
claude mcp add --transport http convops https://mcp.convops.app/ - 2
List the risky moments
Each place your old code asked for approval becomes a step: "send the email", "run the migration", "publish".
- 3
Gate those steps
Mark them as approval gates. The engine refuses to advance until approval is passed, and the advance is written to the audit trail.
- 4
Keep per-call checks where they belong
For approval of individual tool calls, use your AI client's permission settings alongside the workflow gates.
What else do people ask about HumanLayer and ConvOps?
Short answers to the questions that come up most, with the same sources as the table above.
Is the HumanLayer SDK still maintained?
Its public GitHub repository says the code there is pretty much all deprecated and points to the rebuilt HumanLayer product. The company's Research, Plan, Implement framework remains open source.
What is HumanLayer now?
A multiplayer coding-agent IDE and cloud. Tasks group agent sessions, artifacts and worktrees, built-in workflows such as QRSPI guide the work, and teams review design docs with inline comments before code is written.
Does ConvOps approve individual tool calls?
No. ConvOps gates sit on workflow steps and are evaluated by the engine when the agent asks to advance. For approval of a single tool call, use your AI client's permission system.
How does pricing compare?
HumanLayer lists a free Starter plan for up to 3 members, Pro at $100 per user per month and custom Enterprise. ConvOps lists Free, Pro at $79 a month ($66 billed annually) and Business at $299 a month ($249 annually), priced by scale rather than seats, plus Enterprise from $2,500 a month.
Which AI clients does each support?
HumanLayer drives Claude Code, Codex, Copilot and Fireworks inside its own IDE, according to its FAQ. ConvOps works with any MCP client; verified are Claude Code, Claude (web, desktop and Cowork), OpenAI Codex, ChatGPT, Cursor and OpenCode.
Where do these facts come from?
From the official documentation, repository and pricing pages for HumanLayer, checked in October 2026. Anything we could not verify there is left out.
Spotted something out of date? Tell us and we will correct the page and its date.