Should you choose Archon or ConvOps?
Choose Archon when you want a free, local coding pipeline with parallel nodes. Choose ConvOps when the agents you already use must follow a process with sign-off, memory and a record.
choose Archon if
- You want a free, MIT-licensed engine you run yourself, with workflows committed next to the code.
- Your process is a coding pipeline: plan, implement, test, review, open the PR.
- You need deterministic nodes (bash, scripts, tests) and AI nodes in the same graph.
- Parallel nodes, bounded loops and a git worktree per run matter more than which chat client you use.
choose ConvOps if
- The process has to work in ChatGPT, Claude desktop, Cursor and Codex as well as Claude Code.
- Much of the work is not code: content, sales, operations, approvals across teams.
- You want a shared memory so the next session starts knowing what the last one learned.
- You need workspaces for teams, a router for new work and an audit trail with verified humans.
- Example: a nightly support triage grouped in a program with an owner and a goal, which tells you when a run is late, silent or failing.
How do Archon and ConvOps compare?
Side by side, as of October 2026. ConvOps is the operations layer for AI agents. It holds your team's process as workflows with gates, a shared memory, scheduled and unattended runs, programs, a plan board, metrics and a full audit trail. It does not sell or host models. Every Archon cell links to the official source it comes from; ConvOps cells describe the product as it ships today.
| Capability | Archon | ConvOps |
|---|---|---|
| What runs the steps | Archon launches the AI coding agent (Claude, Codex or Pi) for each AI node. Bash and script nodes run without AI. [2] | Your AI client runs each step, or an isolated run ConvOps starts. ConvOps hands it one step at a time and sells no models. |
| Where the process lives | YAML files in .archon/workflows/ in your repo, or ~/.archon/workflows/ for every project, committed with the code. [1] | A workflow in your ConvOps workspace, reached over MCP and changed by asking your AI client. |
| Approval gates | Approval nodes pause the run. Approve or reject from the CLI, the Web UI or a chat platform, and on_reject can route to a fix path. [2] | Gates on any step, evaluated by the engine: an approval, a required metric or a required report. An approval gate holds the step until an explicit approval is given, recorded with the account that gave it, for minutes or days. |
| Memory | The README describes no memory layer in the current version. The earlier task-management and RAG version is archived on a branch. [1] | A shared memory of facts and decisions, recalled across sessions, AI clients and workspaces. |
| Audit trail | Runs, sessions, messages and workflow events are stored in SQLite or PostgreSQL. The Web UI shows an event log per run. [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 | Fire and forget from the CLI, Web UI, Slack, Telegram, Discord or GitHub webhooks. Each run gets its own git worktree. [1] | Attended in your AI session, or unattended through executors and schedules (a cadence or a custom repeat rule, each in its own time zone), including one pod per run on your own Kubernetes. |
| Self-hosting | You run it: a binary, Homebrew, a source checkout or Docker, on your machine or a server. [1] | Hosted at mcp.convops.app, or self-hosted on your Kubernetes with one Helm chart, delivered with Enterprise. |
| Pricing model | Free and open source under the MIT license. You pay your AI provider. [1] | Free $0, Pro $79/mo ($66/mo billed annually), Business $299/mo ($249/mo annually), Enterprise from $2,500/mo. Every process feature on every plan; you pay for scale. Your AI client is billed separately. |
| Client support | Drives Claude Code, Codex and Pi. You start runs from Claude Code (through the Archon skill), the CLI, the Web UI or chat apps. [1] | Any MCP client. Verified: Claude Code, Claude (web, desktop and Cowork), OpenAI Codex, ChatGPT, Cursor and OpenCode. |
| Parallel work and loops | Nodes in the same layer of the DAG run concurrently. Loops repeat until a condition, bounded by max_iterations. [2] | Steps in one workflow run in order; parallel work is child tasks. Loops go back to an earlier step through a decision. |
What is Archon?
Archon is an open-source workflow engine for AI coding agents. You define development processes as YAML workflows, and Archon runs them, each run in its own git worktree.
Its README calls it "the first open-source harness builder for AI coding" and compares it to what Dockerfiles did for infrastructure and GitHub Actions did for CI/CD (Archon README). A workflow is a graph of nodes: prompts, commands, bash, scripts, loops, approvals, waits and sub-runs, joined by depends_on. Nodes in the same layer run concurrently, when: conditions skip branches, and each node can pick its own provider and model (authoring guide).
Archon ships 19 default workflows, such as archon-fix-github-issue and archon-idea-to-pr, plus a web console with run history, event logs and an experimental visual builder. It is MIT-licensed and sends anonymous telemetry that you can switch off.
What is the main difference between Archon and ConvOps?
Who drives. Archon is the orchestrator: it launches the coding agent for each node. ConvOps works two ways: your own AI client pulls one step at a time from a server and reports back, or an executor starts one agent per run in an isolated pod that walks the same steps over MCP.
With Archon, the process is a file in the repository and the engine owns the run. That is a strength for coding pipelines and a constraint outside them: the agents it can drive are Claude Code, Codex and Pi.
With ConvOps, the process lives in a workspace and the run is driven from the client you already use, whether that is Claude Code in a terminal or ChatGPT in a browser. The engine evaluates the gates, persists every step and writes the audit. That is why the same workflow can start in one client and finish in another.
Where is Archon stronger?
In the coding pipeline itself. Archon has primitives ConvOps does not, and it costs nothing to run.
Deterministic nodes
Bash, scripts and tests run as nodes with no AI involved. ConvOps steps are instructions; the engine checks gates and leaves commands to the agent.
Real parallel layers
Nodes with the same dependencies run concurrently. ConvOps runs steps in order and does parallel work as child tasks.
Bounded loops and reject routing
max_iterations caps a loop and on_reject routes a rejected approval to a fix path. ConvOps loops exit on a decision, with no iteration cap.
Worktree isolation built in
Every run gets its own git worktree, so five fixes can run side by side without conflicts.
Free and in git
MIT license, runs on your machine, and the workflow YAML is reviewed and versioned with the code.
Where is ConvOps different?
It holds the process for more than one repo, one client or one person.
Any MCP client
Verified with Claude Code, Claude (web, desktop and Cowork), OpenAI Codex, ChatGPT, Cursor and OpenCode. Each has a setup guide.
Beyond code
Content, sales and operations workflows run on the same engine, with the same gates and record.
A shared memory
Memories and routes are recalled before acting, across sessions and workspaces. See agent memory.
Teams and governance
Workspaces, a router workflow for new work, three roles and an audit row per change with the verified human kept apart.
Managed or on your cluster
Use the hosted server, or run ConvOps on your Kubernetes with one Helm chart, delivered with Enterprise, and one pod per unattended run.
How do Archon concepts map to ConvOps?
Most concepts have a direct equivalent; bash nodes and parallel layers do not. There is no YAML import: you rebuild a workflow by describing it to your AI client.
| Archon | ConvOps | Note |
|---|---|---|
| prompt / command node | Step | The instructions arrive when the step is current. |
| approval node | Approval gate | Engine-evaluated; the step holds until an explicit approval is given, recorded with the account that gave it. |
| loop with until | Decision that invokes a fix sub-workflow and returns | No iteration cap; the exit is a decision. |
| workflow: sub-run | Decision that invokes a sub-workflow | Nests up to five levels; the parent waits. |
| include: | Fragment embedded in many workflows | Edit once, every workflow that embeds it changes. |
| bash / script node | Step that tells the agent to run the command | The engine does not execute commands. |
| parallel layer | Child tasks that share an order | The parent waits for all of them. |
Can you use Archon and ConvOps together?
You can, but pick one engine to own each process. Two engines holding the same pipeline means two places to change it.
A split that works: Archon runs a repo-local coding pipeline, and ConvOps holds the process around it, from intake and scope approval to release sign-off and the announcement. Because Archon is started from Claude Code through its skill, a ConvOps step can simply say "use archon to run archon-idea-to-pr for this task". That is a pattern, not an integration we ship.
one ConvOps workflow, with Archon inside one step
How do you move an Archon workflow to ConvOps?
Connect a client, describe the workflow, then place the gates and the fix loop.
- 1
Connect ConvOps
In Claude Code, one line. Codex, Cursor, OpenCode, ChatGPT and Claude Cowork take the same URL; each has a setup guide.
claude mcp add --transport http convops https://mcp.convops.app/ - 2
Describe one workflow
Paste the Archon YAML and ask: "Create a ConvOps workflow with these steps." Your AI client builds it step by step.
- 3
Place gates and loops
Turn approval nodes into approval gates. Turn test-and-fix loops into a review decision that invokes a fix sub-workflow and returns to Verify.
- 4
Run it in the client you use
Create a task. The workflow attaches, and your AI client walks it one step at a time.
What else do people ask about Archon and ConvOps?
Short answers to the questions that come up most, with the same sources as the table above.
What is the difference between Archon and ConvOps?
Archon is a free, open-source engine that runs YAML coding pipelines, launching Claude Code, Codex or Pi in isolated git worktrees on your machine. ConvOps is the operations layer for AI agents: it holds the process on a server that any connected AI client follows one step at a time, with gates, a shared memory, schedules and an audit trail, for coding and non-coding work.
Is Archon free?
Yes. Archon is open source under the MIT license, and you run it yourself. You pay for the AI provider it drives, such as your Claude or Codex plan. ConvOps also has a free plan with every process feature included, and you keep using the AI client you already pay for.
Does ConvOps launch Claude Code the way Archon does?
Not in attended mode. Your AI client connects to ConvOps and pulls one step at a time. For unattended runs, an executor starts the agent, one isolated pod per run on ConvOps Cloud or on your own Kubernetes. Either way the agent works one step at a time, and the record shows what each step produced.
Can I import Archon YAML into ConvOps?
No. ConvOps has no YAML import. You rebuild a workflow by describing it to your AI client, which creates the steps, gates and decisions over MCP. Pasting the Archon file into the conversation as a reference is a quick way to start, and you then adjust each step's instructions by asking.
Does ConvOps have bash or script nodes?
No. A ConvOps step is instructions for the agent or a person. The engine evaluates gates against real state, such as approval or finished child tasks. Commands are run by the agent the step instructs. So a step that needs a script says which one to run, and the agent runs it in its own environment and records what it found when it advances.
Which is better for a solo developer on one repository?
Often Archon. If the whole process is a coding pipeline in one repo and you work in Claude Code, Codex or Pi, a free local engine with worktrees and parallel nodes is a strong fit. ConvOps pays off when the process crosses clients, people or non-coding work.
Where do these facts come from?
From the official documentation, repository and pricing pages for Archon, 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.