Why protoAgent
Most agent frameworks hand you everything and dare you to delete what you don't want. protoAgent inverts that: a small core you actually understand, and everything else as a plugin — built to run distributed and orchestrate a fleet over A2A.
Bare-bones core, opt-in everything
You start with a small, legible core and add only what you need — tools, skills, subagents, workflows, console views, memory backends — as plugins. No inheriting a pile of use-cases you'll never use, then fighting to remove them.
A2A-native, built for fleets
Every protoAgent is a first-class A2A 1.0 server and can fan out to other A2A endpoints (delegate_to). It's designed to run distributed and be orchestrated as a fleet across your machines — not as one monolith on one box.
Local-first, model-agnostic
Powered by your models through an OpenAI-compatible gateway — local 30B, a hosted frontier model, whatever you point it at. Not a coding-agent competitor: it orchestrates work (and can drive any coding agent over ACP).
protoAgent vs Hermes & OpenClaw
✓ yes · ~ partial / via add-ons · — not a focus
| protoAgent | Hermes | OpenClaw | |
|---|---|---|---|
| Foundation | LangGraph + A2A 1.0 | own runtime | own gateway |
| Center of gravity | A2A orchestration substrate | self-improving personal agent | multi-channel chat gateway |
| Core stance | minimal core, opt-in plugins | feature-rich built-ins | feature-rich built-ins |
| Extend without forking | ✓ git-URL plugins | ✓ plugins | ✓ skills/plugins |
| Agent-to-agent | ✓ A2A 1.0 (HTTP/JSON-RPC), shipped | ~ proposed (#514) | ~ encrypted relay |
| OpenAI-compatible endpoint | ✓ | ✓ | ~ |
| Pluggable memory / vector store | ✓ | ✓ providers | ~ markdown |
| Built-in cost/latency telemetry + CSV export | ✓ | — (local-only by design) | — |
| Langfuse tracing | ✓ first-class | ✓ plugin | ~ |
| Operator console + custom plugin dashboards | ✓ (+ native desktop) | ~ | ~ |
| Chat-channel breadth | ~ Discord/Telegram/Slack | ✓ broad | ✓ 50+ (widest) |
| Delegate to coding agents | ✓ ACP | ~ | ✓ |
| Autonomy (scheduler / heartbeat) | ✓ + goal mode & verifiers | ✓ | ✓ heartbeat |
| Sandboxing / NVIDIA | ✓ OpenShell + egress fence | ~ | ✓ NemoClaw |
| License | MIT / fork-to-ship | MIT | MIT |
Hermes (NousResearch) and OpenClaw are capable, MIT-licensed, local-first agents too — this maps where protoAgent's design priorities differ, not a claim that it does more. They each lead in places (OpenClaw's 50+ chat channels, Hermes's self-improving memory). Researched June 2026; these tools move fast — corrections welcome.
The features that make the difference
A2A 1.0 as the native interface
protoAgent IS a spec-compliant A2A 1.0 server (HTTP/JSON-RPC, agent card, extensions) — shipped, not proposed. A hot-swappable delegate registry fans a sub-task out to other a2a / openai / acp endpoints. The whole design assumes a fleet: agents discovering and delegating to each other over the open standard.
Built on LangGraph
The runtime is LangGraph — durable checkpointing, the tool/subagent loop, and the LangChain ecosystem underneath. You're extending a graph, not a bespoke loop, so what you learn here transfers.
A lean core, not a bundle
Hermes and OpenClaw ship rich built-ins (browser automation, voice, 50+ chat channels). protoAgent ships a small legible core and you add the rest as git-URL plugins — pinned in a lockfile, removable cleanly, shareable as repos. You opt in to surface area instead of opting out of it.
An operator console, not just chat apps
Where the others live inside your chat apps, protoAgent centers on a dedicated React console (+ a native desktop app) where a plugin can add its own left-rail dashboard — the SpaceTraders Fleet view, a Quant Desk — without a rebuild. Manage a fleet from one surface.
Built-in cost & latency telemetry
A per-turn telemetry store (cost, p50/p95 latency, tokens, cache-hit, by-model) with CSV export and a retention guardrail — plus first-class Langfuse + Prometheus. Some agents deliberately keep zero telemetry; protoAgent treats cost/latency visibility as core.
Goals with testable outcomes
Beyond a scheduler: goal mode with pluggable verifiers (a goal is met when a check passes, not when the model says so) and monitor goals for long-horizon, externally-driven targets — e.g. a background loop grinding toward 1M credits, evaluated out-of-band.
Drive any coding agent (ACP)
Delegate a real coding job to protoCLI, Claude Code, Codex, or Gemini CLI over the Agent Client Protocol. protoAgent orchestrates coding agents rather than competing as one.
Headless, API-first
Run it with no UI (--ui none): A2A + an OpenAI-compatible /v1 endpoint + /metrics. Point any OpenAI client or agent fleet at it. The console is optional, not the product.
Stress-tested by building on it
We don't just ship the core — we build real agents on it and contribute the patterns back as plugins. A SpaceTraders agent runs an OODA loop on the scheduler (observe every 20 min, escalate to a strategist subagent hourly) toward a standing goal of 1,000,000 credits — with a Fleet dashboard, entirely as plugins that never touch core. A protoTrader Finance plugin adds market data, backtesting, and a gated paper broker. ORBIS (voice companion) connects over A2A to control it by voice. Same substrate, wildly different agents.