BerriAI

Query spend and token activity on a live LiteLLM proxy. Shows daily usage broken down by user, team, org, tag, job, or model. Use when the user wants to see costs, token counts, request volume, or job-level attribution for a given date range.

Generate a new API key on a live LiteLLM proxy. Asks for alias, scope (user/team), budget, models, and expiry, then calls POST /key/generate. Use when the user wants to create, generate, or provision an API key on a LiteLLM proxy instance.

Add a new model to a live LiteLLM proxy. Walks the user through picking a provider, entering the deployment name and credentials, calls POST /model/new, then test-calls the model to confirm it routes correctly. Use when the user wants to add, register,…

Create a new team on a live LiteLLM proxy. Asks for team name, budget, and allowed models, then calls POST /team/new and shows the result. Use when the user wants to create a new team, set up team budgets, or configure model access for a team on the proxy.

Delete a model from a live LiteLLM proxy. Asks for the model name or model_id and confirms before calling POST /model/delete. Use when the user wants to remove, delete, or unregister a model from a LiteLLM proxy instance.

Delete an AI agent from a live LiteLLM proxy. Ask for the agent_id and confirm before calling DELETE /v1/agents/{agent_id}.

Delete one or more API keys from a live LiteLLM proxy. Ask for the key(s) and confirm before calling POST /key/delete.

Delete an MCP server from a live LiteLLM proxy. Ask for the server_id and confirm before calling DELETE /v1/mcp/server/{server_id}.

Organize database repositories and proxy dependency wiring, manage connection ownership, or port Prisma operations to Rust and SQLx without expanding migration scope

Keep secrets out of Debug output in litellm-rust with SecretValue and veil. Use when a struct or enum holds a password, token, key, credential URL or credential provider, when writing or reviewing a Debug impl, or when a type with secrets gets logged or traced

Define, wrap, convert, classify, report or test Rust error types in litellm-rust with thiserror. Use when adding an error type or variant, mapping one crate's error into another, turning an error into a gateway response or Python exception, deciding retry or…

Add or change an HTTP route, extractor, response envelope, stream or router mount in the litellm-rust gateway crates (crates/gateway and crates/gateway-*). Use when adding an endpoint, parsing a request body, choosing an error envelope for an API surface,…

Define or refactor Rust string-valued enums and their Serde adapters in litellm-rust, using Strum and serde_with while preserving parsing, wire values, and schemas

Add or change Rust diagnostic tracing in litellm-rust, including route spans, subscriber layers, and Python logger delivery

Set up a self-hosted LiteLLM proxy, the OpenAI-compatible LLM gateway, on this machine, from install through provider keys, config.yaml, a master key, optional Postgres and Admin UI, and a verified chat completion. Use when the user asks to set up LiteLLM,…

Knowledge and utilities for creating animated GIFs optimized for Slack. Provides constraints, validation tools, and animation concepts. Use when users request animated GIFs for Slack like "make me a GIF of X doing Y for Slack."

Build or refine Moyai's native React frontend using its established navigation, typography, forms and interaction patterns. Use for native/web and shared frontend changes or UI reviews, not backend-only work.

Implement or fix code in a repository, including setup, tests, and a reviewable change.

Keep the Python API and the native/web frontend type-safe end to end (Pydantic → OpenAPI → hey-api generated client). Use whenever adding or changing a FastAPI route, a request/response payload, or a frontend call to /api.

Read and operate native desktop apps on the shared computer through Moyai's desktop tools.

Answer "do we support X?" for a linked provider doc, feature, or API by checking the code, then reply with a TL;DR verdict, what it would take, and an offer to do it.

Deliver implemented code as a regular pull request with verification and working-demo evidence.

Investigate a technical question using current primary documentation and source evidence.

Review code or a pull request for concrete correctness, security, recovery, and regression risks.

Triage a GitHub pull request for BerriAI/litellm and decide whether it is ready for human review. Classifies each failing check as PR-related vs infra/pre-existing, factors in Greptile score and CircleCI presence, and emits a thumbs-up/thumbs-down verdict…

Iteratively fix a BerriAI/litellm PR until the litellm-bot reviewer returns a 5/5 READY verdict. Posts the PR to the bot (via Slack @-mention or its /chat/api endpoint), parses the merge-confidence card, fixes every blocker the card and drilldown call out,…

Karpathy-style senior-engineer pre-merge review for a litellm PR as if it ships to an AI gateway at 10k+ RPS on a large codebase with many flows. Answers whether the change is safe to merge, what unintended consequences could appear under load, and whether…

Review a BerriAI/litellm PR for conformance with the repo's documented and de-facto code patterns. Pulls patterns from docs/my-website/docs/ first, then from sibling files in the diff's directories. On conflict, docs win and contradicting code is flagged as…

Standard procedures for a coding agent — memory hygiene and PR workflow.

Audits all Markdown files in a repository for internal inconsistencies, contradictions, stale CLI commands/flags, and broken cross-references. Returns a structured findings report. Use when asked to check, audit, or verify docs for correctness or consistency.

Scans a GitHub pull request for security vulnerabilities including OWASP Top 10 issues, hardcoded secrets, and dependency CVEs. Returns a structured JSON findings report. Use when asked to review, audit, or scan a PR for security issues.

Produce the secret-free Codex config to distribute to a whole team via MDM, and plan the gateway deployment behind it. Use for questions about rolling Codex out company-wide, shipping config.toml to every laptop, per-user keys, or centralizing spend.

Stand up the LiteLLM gateway and point the Codex CLI at it. Use when someone wants Codex running against a self-hosted gateway instead of calling OpenAI directly, or says "set up the stack", "wire Codex to LiteLLM", or "run the setup".

Diagnose a Codex CLI that won't reach its LiteLLM gateway: connection refused, 401s, model not found, hanging streams, or requests bypassing the gateway. Use when setup failed, doctor.sh reports a failure, or Codex errors after previously working.

Read or write dashboard URL query state (tabs, filters, pagination, deep links, demo flags) with nuqs instead of raw history or URLSearchParams

Iterate on a pull request with LiteReview until its current commit earns 5/5, then request a hosted preview for the human reviewer. Use when asked to run the LiteReview fix-and-review loop.

Design, implement, and review LiteLLM Memory screens using the project's visual hierarchy, interaction patterns, product language, and privacy semantics. Use for LiteLLM Memory UI changes and extensions, including setup, activity, chat, settings, and…

Set up, pause, resume, inspect, or remove the LiteLLM Lens integration for local Codex chats.