Skip to main content

Tips & Best Practices

These are quick improvements for everyday Anyy use. Scan the headings and apply only the sections relevant to your current surface and task.

Get better results​

State the outcome and constraints​

Describe what “done” means, the important boundary, and the evidence you expect. “Compare these proposals and recommend one under a $500 budget” gives Anyy more useful direction than “look at these.”

Provide precise context up front​

Include relevant paths, error text, time range, audience, and required output format in the first message. Use Context References when the source already exists:

Review @file:workspace/proposal.md and return a decision table with risks and open questions.

Separate durable and temporary information​

Ask Anyy to remember stable preferences and facts. Keep one-off instructions, copied sources, task progress, and secrets out of long-term memory.

Use Skills for repeated procedures​

If the same multi-step workflow recurs, install or write a Skill rather than pasting a long procedure into every chat. Inspect third-party instructions before enabling them, and remember that a Skill does not grant tool authority.

Choose the right boundary​

Use one session for one sustained topic. Start a new session when the goal, Role, model, or context changes substantially.

Use a Role for a different assistant identity that shares the same Profile. Use a separate Profile when memory, credentials, sessions, or trust domains must not mix.

Name the session early

A clear title makes session search and later resumption more useful. Dashboard can keep separate ordinary and background sessions visible.

Work efficiently in Dashboard and TUI​

  • Use Shift+Enter for a newline in Dashboard.
  • In the TUI, Shift+Enter, Alt+Enter, or Ctrl+J inserts a newline.
  • Type / to browse commands available to the current surface and runtime.
  • Use Steer for a correction meant to redirect active work; use the ordinary queue for a separate next task.
  • Interrupt only after checking whether the current delay is a provider call or long-running tool.

Resume a known session from the terminal with:

anyy --resume SESSION_ID

Control cost and context​

Use a model that fits the task​

Reserve expensive or high-reasoning routes for work that benefits from them. Use the installed provider catalog and model verification rather than assuming a model ID works on every endpoint.

Start fresh after large changes​

Changing model or reasoning effort can invalidate prompt cache reuse and increase input-token cost. Start a new session when a clean context is more valuable than continuity.

Keep attached context focused​

Use a file line range, a smaller set of references, or session search instead of attaching entire directories or many long sources. Anyy enforces finite reference and tool-result budgets.

Let Anyy use tools without granting too much​

Ask for the result, then let Anyy choose from enabled tools. Keep approval mode on so state-changing commands and file edits pause when required.

Review the proposed target, commands, expected impact, validation, and rollback—not just the approval button label.

Yolo is a trust decision

Use yolo only for a narrow session or schedule whose inputs, toolsets, working directory, and external effects you understand. Absolute safety interlocks remain, but ordinary changes may run without asking.

Configure integrations one at a time​

For a new Channel, MCP server, browser backend, or media route:

  1. configure it;
  2. run its test, probe, status, or doctor flow;
  3. perform one small real task;
  4. inspect delivery and audit evidence; and
  5. only then enable unattended use.

This keeps authentication, reachability, permissions, and routing failures distinguishable.

Design reliable automations​

  • Put one clear outcome in each schedule.
  • Use an explicit Role, toolsets, execution mode, and delivery target.
  • Test with Run now before relying on the next due time.
  • Prefer approval mode until the behavior is stable.
  • Review history and receipts after the first unattended run.
  • Disable a failing job before editing several variables at once.

Keep Anyy recoverable​

Create a backup before broad model, Channel, Profile, Skill, or MCP changes. Verify the archive and occasionally restore into a new directory; a backup that has never been restored is only an attempt.

Run this basic health sequence after material changes:

anyy status
anyy doctor
anyy gateway status

Security habits​

  • Never paste API keys into chat, memory, Soul, Skills, screenshots, or shell arguments.
  • Treat Dashboard token URLs as temporary credentials.
  • Review Channel pairing identities before approval.
  • Keep MCP include lists narrow.
  • Read third-party SKILL.md files and supporting scripts.
  • Rotate a credential at its provider first when exposure is suspected.

For a problem that persists, continue with FAQ & Troubleshooting.