ThumbGate vs OpenMono
OpenMono (StartupHakk, AGPL-3.0) is a local-first coding agent: llama.cpp on your box, Docker sandbox, 12-step permissions pipeline, doom-loop abort at a fixed 3× repeat. ThumbGate is not another agent. It is the learning firewall that makes those controls persist across sessions.
The short answer
Not a substitute. OpenMono answers “run a capable agent with zero token meter and a hard sandbox.” ThumbGate answers “did this repo already teach us that this tool sequence wastes the run?” Pitching them “you need guardrails” would be false. The gap is persistence: run 1 and run 500 are identical unless an operator hand-edits config.
Comparison table
| Dimension | OpenMono | ThumbGate |
|---|---|---|
| What it is | Local coding agent (.NET 10 + llama.cpp + Docker) | Infrastructure firewall for any coding agent |
| Question it answers | Can I run an agent with no cloud meter? | Will this tool call repeat a known-bad pattern? |
| Enforcement | 12-step pipeline; nothing bypasses it; sandbox mount is the blast radius | PreToolUse hook: allow / warn / deny before execute |
| Loop control | Fixed 3× repeated tool sequence abort | Learned sequences from this repo’s wasted runs |
| Turn budgets | Explore 15 / Plan 10 / Coder 30 / Verify 20 — same everywhere | Measured per-repo distributions when the adapter is wired |
| Cross-session memory | Per-run; operator edits config | Feedback → lesson → prevention rule → next-call block |
| License | AGPL-3.0 | Permissive public runtime; adapter talks over their existing interface |
| Pricing | $0 inference after one-time hardware setup | Free local CLI; self-serve Pro at $19/mo for unlimited captures |
When to use OpenMono
- You want the agent and the model on your hardware with no per-token bill.
- You already like their Docker workspace mount and specialist sub-agents.
- You need a TUI / VS Code / mobile client that is the agent, not a gate around someone else’s agent.
When should I use ThumbGate?
- The same destructive or wasteful tool sequence keeps recurring across OpenMono runs.
- You want Monday’s thumbs-down to change Tuesday’s pipeline without hand-editing budgets.
- You run OpenMono next to Claude Code / Codex / OpenCode and want one lesson store.
Running both
OpenMono’s pipeline is the chokepoint. ThumbGate maps onto it; it does not replace the 12 steps. Integration proposal: StartupHakk/OpenMonoAgent.ai#129. The ThumbGate-side adapter contract is a separate in-flight PR and is not claimed shipped until it merges.
Hosted Enterprise capabilities are not generally available.
FAQ
Is OpenMono a ThumbGate competitor?
No. OpenMono is the runtime. ThumbGate is the learning gate on the tool call.
Does OpenMono already have guardrails?
Yes — sandbox, pipeline, fixed 3× doom-loop, turn budgets. They do not learn from last session.
When should I use ThumbGate?
When a failure should become a rule instead of a wiki note. Free local evaluate stays free.