Compare
How Spettro actually compares.
Against single-agent CLIs, editor chat panes, and Claude.ai plans - capability by capability, number by number.
Spettro vs single-agent CLIs vs editor chat
Single-agent terminal assistants and editor chat panes collapse the whole job into one prompt. Spettro splits planning, coding, review, docs, git, and testing into roles you can see - with explicit handoffs and an approval flow between them.
| Spettro | Single-agent CLI | Editor chat | |
|---|---|---|---|
| Agent model | Multi-agent roles with explicit handoffs | One agent, one prompt | Chat pane inside the editor |
| Autonomous goals | /goal loop with stall detection | Not available | Not available |
| Agent swarm | Ultra mode - parallel sub-agents | Not available | Not available |
| Approval flow | Ask-first / restricted / yolo modes | Per-tool prompt | Per-tool prompt |
| Provider routing | Anthropic, OpenAI-compatible, local (LM Studio, Ollama) | Usually vendor-locked | Usually vendor-locked |
| OS sandboxing | Seatbelt / Landlock / Mandatory Integrity Control | Rarely available | Not available |
| Sessions | Persistent sessions, /resume, /compact | Chat history | Chat history |
| Extensibility | Skills, hooks, custom slash commands, per-repo manifests | Global config only | Workspace settings |
| Memory | User-scope and project-scope /memory | Varies | Varies |
| Remote control | HTTP/SSE control plane, Telegram relay, ACP | Not available | Not available |
| License | Open source, GPL 3.0 | Varies | Usually proprietary |
Spettro vs Claude.ai plans
Spettro runs on a transparent credit budget - you always know exactly how much you have left. Claude.ai measures usage by session capacity, with no hard token number and no BYOK.
| Spettro | Claude.ai | ||||
|---|---|---|---|---|---|
| Pro€20/mo | Max€150/mo | Pro$20/mo | Max 5×$100/mo | Max 20×$200/mo | |
| Monthly budget (on SuperFast) | ~200M tokens | ~1.5B tokens | session limit | 5× sessions | 20× sessions |
| Models available | 9 models | 9 models | 2 models | 2 models | 2 models |
| Unlimited fast model | — | — | — | ||
| Bring your own key | — | — | — | ||
| CLI & agentic tools | |||||
Where Spettro is genuinely different
Plan, code, review, docs, git, test, and explorer agents with explicit handoffs - not one prompt pretending to be a team.
Goal mode runs unattended across iterations; Ultra mode fans a task out to a swarm of parallel sub-agents.
Kernel-level confinement on all three platforms, set by the operator and invisible to the model.
Open source under GPL 3.0, BYOK on every plan, and local model support - no vendor lock-in at any layer.
Comparison questions
What makes Spettro different from Claude Code?
Claude Code is a single-agent first CLI locked to Anthropic models. Spettro splits work across explicit agent roles (plan, code, review, docs, git, test, explorer), adds goal mode for autonomous runs and Ultra mode for parallel agent swarms, and routes to any provider: native Anthropic, OpenAI-compatible APIs, or local endpoints like LM Studio and Ollama.
Is Spettro cheaper than a Claude.ai subscription?
Spettro plans start at €5/month with a transparent token budget - Pro is €20 for ~200M tokens and Max is €150 for ~1.5B tokens, with an unlimited fast model after quota. Claude.ai measures usage in session capacity with no hard token number and no bring-your-own-key option. On the Spettro Free plan you pay nothing and use your own API key.
How does Spettro compare to editor chat like Cursor or Copilot?
Editor chat panes embed one assistant in a proprietary editor. Spettro is a terminal-native, open-source (GPL 3.0) runtime with kernel-level OS sandboxing, approval flow, persistent sessions, and per-repo agent manifests - and it integrates with ACP-capable editors when you want it inside one.
Can I use Spettro without a subscription at all?
Yes. Spettro is open source and the Free plan is fully functional with your own API key (BYOK) - every feature is included, with usage limited only by your key's provider.
Does Spettro support local models?
Yes. Spettro connects to local OpenAI-compatible endpoints such as Unsloth, LM Studio or Ollama, so you can run agents entirely against local models - something most vendor-locked assistants cannot do.