Magellan can use context from ShellYard tools — terminal output, HTTP responses, database results, diagnostics, config diffs, docs, and IPAM — to explain what happened and help document the fix. BYO Anthropic, OpenAI, Gemini, or Ollama. You pick what's attached. The operator runs every command.
Read what you shared and tell you what it means. No actions.
Terminal output from demo@core-switch.lab.internal:2201 (last 7 lines) auto-attached.
Magellan reads selected output from any ShellYard tool — terminal, HTTP response, database query, packet capture, switch config, log line — and reasons about it inside the same workspace where you'll run the fix. No copy-paste into a browser tab. No leaking the wrong client's data into someone else's chat history.
Scenes from the work
Scene 01
Read what just happened
Capture packets on a flaky link, watch retransmits stack up, then send the snapshot to Magellan. It reads the timing pattern and tells you whether you're looking at a congested uplink, a duplex mismatch, or an MTU black hole — grounded in the bytes you just captured, not a generic answer.
Scene 02
Draft the fix
Paste a Cisco IOS startup-config. Magellan reads the VLAN layout, spots the orphaned port-channel, and drafts the no-shut + interface-range commands. You read the draft, edit if you want, and run it. Nothing fires until you say so.
Scene 03
Hand off to a runbook
After the page closes, pipe the full session into Magellan and ask for a runbook. It pulls out the steps you actually ran, the commands that worked, and the dead ends — written as a runbook your on-call peer can follow tomorrow.
BYO key
ShellYard does not need to be the model. ShellYard is the interface between your infrastructure work and the model you choose — same flow, four providers.
Bring your OpenAI API key. Choose the model you trust for the work — Magellan is the interface, not the inference.
Same flow with Anthropic — paste your key, pick a Claude model, work with it inside the terminal.
Same flow with Google — paste your AI Studio key, pick a Gemini model, run it inside the terminal.
Point Magellan at a local Ollama (or LM Studio / OpenAI-compatible) endpoint and keep prompts and responses on your network.
On Free, your provider key lives in a local AES-256-GCM-encrypted SQLite vault on disk, sealed by your OS keychain — it never leaves the device. Prompts and responses pass directly between your client and your chosen provider; ShellYard does not meter, mark up, or proxy AI usage.
Operator control
Magellan moves from reading to action only when an operator explicitly switches it there. Tool-call access is protocol-level gated — the model can't emit a command in Draft mode, because the command-running tools aren't even shown to it until Execute.
Reads attached output, tells you what it means.
Proposes the next diagnostic step.
Writes the command. Nothing runs.
You read the draft, edit if needed.
You run it. Operator confirms every step.
Three-layer enforcement details — protocol, frontend executor, backend binding — live in the five-stage flow doc.
Ask Magellan
Each one starts with context Magellan already has — the active terminal session, the last HTTP response, a config backup, the Space's docs. You don't paraphrase. You point.
> Explain this failed API request.
> Summarize this terminal session into a runbook.
> Compare these config backups and identify what changed.
> Search this Space for anything related to DHCP failures.
> Create an incident report from this troubleshooting session.
What you can count on
You control what context is sent.
Magellan only receives the snapshots you attach or authorize from ShellYard tools.
Secrets are not sent by default.
Raw credentials, private keys, and password fields are excluded from attached context.
Bring your own AI provider.
Anthropic, OpenAI, Gemini. Your key, your account, your billing relationship. ShellYard takes no markup.
Use a local model if you want one.
Ollama or any OpenAI-compatible endpoint — point Magellan at it and your context never leaves the machine.
Multi-tenant by default
Magellan operates inside the active Space. Switch to a different client and the active-tool snapshot store flushes — Acme's terminal output doesn't haunt a question about Beta's network. On Enterprise, each Space gets its own AWS KMS key, so cryptographic erasure on client offboarding takes the AI-generated notes with it.
Questions we hear
Magellan is BYO key on every tier. Prompts go directly from your machine to Anthropic, OpenAI, Google, or your local model — no ShellYard server in the request path. Free is enough to try the full flow with your own provider. Team adds Shared Spaces with per-Space CMK and Group-scoped credentials; Enterprise adds per-Space audit-log filtering and larger storage caps.
