# Documentation

- [CrabDash Gateway API](/docs/crabdash/gateway-api): Use CrabDash's local OpenAI-compatible and Anthropic-compatible endpoints from any app, SDK, or CLI tool.
- [Getting Started with CrabDash](/docs/crabdash/getting-started): Download CrabDash, add your first LLM provider key, and route requests through your personal local gateway.
- [CrabDash Local Models](/docs/crabdash/local-models): Run ~700 MLX community models locally on Apple Silicon — no Docker, no Ollama, no separate process.
- [CrabDash Free Provider Keys](/docs/crabdash/provider-keys): Get free LLM API keys to use with CrabDash — no credit card, no waitlist.
- [Architecture](/docs/crabllm/architecture): Design principles, workspace layout, and request flow through the CrabLLM gateway.
- [Configuration](/docs/crabllm/configuration): Full reference for crabllm.toml — providers, virtual keys, aliases, pricing, extensions, and storage.
- [Docker](/docs/crabllm/docker): Run CrabLLM in Docker with persistent storage and manage it with crabctl from the host.
- [Getting Started](/docs/crabllm/getting-started): Install CrabLLM, configure providers, and send your first request in under a minute.
- [Introduction](/docs/crabllm/introduction): High-performance LLM API gateway in Rust — one API format, many providers, sub-millisecond overhead.
- [Libraries](/docs/crabllm/libraries): Standalone Rust crates for local inference — MLX (Apple Silicon) and llama.cpp (cross-platform).
- [Agents](/docs/crabtalk/agents): How CrabTalk agents work — configuration, system prompts, delegation, and scoping.
- [Auth](/docs/crabtalk/auth): Manage providers and MCP servers with the interactive auth TUI.
- [Commands](/docs/crabtalk/commands): CrabTalk commands — cargo-style subcommand dispatch for telegram, search, and custom tooling.
- [Configuration](/docs/crabtalk/config): Configure CrabTalk — config.toml, models, MCP servers, and agents.
- [Cron](/docs/crabtalk/cron): Daemon-level cron — run skills on a schedule, even when no client is connected.
- [Events](/docs/crabtalk/events): Real-time event stream — monitor every agent, every tool call, every session from one connection.
- [Hub](/docs/crabtalk/hub): Discover and install MCP servers, agents, and commands from the CrabTalk hub.
- [Installation](/docs/crabtalk/installation): Install CrabTalk on macOS, Linux, or Windows with one command, or build from source with Cargo.
- [MCP Servers](/docs/crabtalk/mcp-servers): Connect external tool servers to CrabTalk — stdio or HTTP, auto-discovered, hot-reloadable.
- [Memory](/docs/crabtalk/memory): Local-only agent memory — remember, recall, and forget, all on your machine.
- [Protocol](/docs/crabtalk/protocol): The wire protocol between CrabTalk clients and the daemon — protobuf over UDS or TCP.
- [Providers](/docs/crabtalk/providers): How CrabTalk routes model requests — API standards, provider config, and supported providers.
- [Quickstart](/docs/crabtalk/quickstart): Run your first CrabTalk agent in under a minute — start the daemon and chat.
- [REPL](/docs/crabtalk/repl): Interactive chat with your agent — slash commands, tab completion, streaming output.
- [Sessions](/docs/crabtalk/sessions): Persistent conversations — append-only JSONL, auto-compaction, crash-safe.
- [Skills](/docs/crabtalk/skills): Prompt-level behavior for agents — Markdown files with YAML frontmatter, discoverable and composable.
- [Tutorials](/docs/crabtalk/tutorials): Step-by-step guides for setting up and extending CrabTalk.
- [Authentication](/docs/crabllm/features/auth): Virtual API keys with per-key model access control, usage tracking, and no-auth mode.
- [Extensions](/docs/crabllm/features/extensions): Request pipeline hooks — caching, rate limiting, usage tracking, budget enforcement, and logging.
- [Management](/docs/crabllm/features/management): Dynamic provider and key management via crabctl CLI, admin API, and OpenAPI docs.
- [Routing](/docs/crabllm/features/routing): Weighted random selection, exponential backoff retry, automatic provider fallback, and timeouts.
- [Storage](/docs/crabllm/features/storage): Pluggable storage backends — memory, SQLite, and Redis for extension state persistence.
- [Streaming](/docs/crabllm/features/streaming): SSE streaming across all providers — proxied without buffering, with per-chunk extension hooks.
- [Anthropic](/docs/crabllm/providers/anthropic): Configure Anthropic Claude — full request/response translation including system messages, tool calling, and streaming.
- [Azure OpenAI](/docs/crabllm/providers/azure): Configure Azure OpenAI — deployment-based URL rewriting and API version handling.
- [AWS Bedrock](/docs/crabllm/providers/bedrock): Configure AWS Bedrock — Converse API translation with SigV4 request signing.
- [Google Gemini](/docs/crabllm/providers/google): Configure Google Gemini — full translation including system instructions, function calling, and streaming.
- [Ollama](/docs/crabllm/providers/ollama): Configure Ollama for local model inference — no API key, OpenAI-compatible pass-through.
- [OpenAI](/docs/crabllm/providers/openai): Configure OpenAI and any OpenAI-compatible API — Groq, Together AI, vLLM, and custom endpoints.
- [Providers](/docs/crabllm/providers/overview): Supported LLM providers, translation types, and endpoint support matrix.

# Blog

- [How agents call agents](/blog/agent-calling-patterns): We surveyed eight frameworks to map who can call whom in agent teams — upward, lateral, recursive. Most are more restricted than expected.
- [What should an agent capability bench test?](/blog/agent-capability-benchmarks): A survey of existing agent benchmarks and 120+ questions we think a practical capability bench should answer.
- [When agents share a channel](/blog/agent-loop-prevention): We tested what happens when two agents join the same group chat. Most frameworks have no built-in answer for the infinite loop that follows.
- [How developers configure their AI agents](/blog/agent-prompt-systems): A survey of CLAUDE.md adoption, MCP servers, skills, and hooks — and why your prompt system decays predictably from MVP to production.
- [Sandboxing AI agents: beyond Docker and WASM](/blog/agent-sandbox-permissions): How Claude Code, Cursor, Codex, and Devin gate bash access — and why Docker, WASM, and OS sandboxes are all solving the wrong layer.
- [Async compaction: the race conditions nobody talks about](/blog/async-compaction-challenges): We examined how frameworks handle background compaction. Most avoid it entirely. The ones that try it have documented race conditions and silent data loss.
- [L0 to L5: where AI agents land on the autonomy spectrum](/blog/autonomy-spectrum): A survey of how AI agents are shifting from human-approved to human-supervised — autonomy levels, product positioning, and what it means for agent runtimes.
- [Build a Gateway](/blog/build-gateway): Create a custom gateway — connect any messaging platform to CrabTalk agents.
- [Build an MCP Server](/blog/build-mcp-server): Build an MCP server as a crabtalk command — start/stop, service management, and auto-discovery.
- [Built-in agents: what ships in the box](/blog/builtin-agents-landscape): We surveyed ten AI coding products to catalog their built-in agents. Three have real multi-agent delegation. The rest ship modes.
- [Built-in tools: what your agent can reach](/blog/builtin-tools-landscape): We cataloged the built-in tools in ten AI coding products. Some ship 5, some ship 20+. The differences reveal what each product thinks an agent should do.
- [Built-in web search: no API keys, no setup](/blog/builtin-web-search): How we gave every CrabTalk agent web search and page fetching — with multi-engine consensus ranking, zero API keys, and zero configuration.
- [Context compaction in agent frameworks](/blog/context-compaction): We surveyed eight frameworks to understand who compacts what in multi-agent systems. The answer: each agent handles its own, and nobody coordinates.
- [Hello from CrabTalk](/blog/hello-crabtalk): Introducing CrabTalk — a local-first autonomous AI agent daemon built in Rust.
- [Hermes Agent: what Nous Research built](/blog/hermes-agent-survey): We examined Hermes Agent's architecture — from Atropos RL training to persistent skill documents. Here's how it works and where it fits.
- [Hermes memory: five layers, one learning loop](/blog/hermes-memory-system): We examined Hermes Agent's five-layer memory system — procedural skills, Honcho user modeling, FTS5 search — and asked what it costs to remember this much.
- [Mem0: what three memory scopes actually cost](/blog/mem0-memory-architecture): We examined Mem0's extraction pipeline, conflict resolution, and benchmark claims. Smart memory management is real — but most agents don't need three scopes.
- [MiroFish: swarm prediction through social simulation](/blog/mirofish-swarm-prediction): We examined MiroFish's architecture — from OASIS swarms to GraphRAG extraction. How it predicts the future and where the approach breaks down.
- [Why multi-agent workflows fail in production](/blog/multi-agent-coordination): A survey of real coordination failures in Claude Code, Cursor, Devin, and the broader ecosystem — and what they reveal about agent architecture.
- [Multi-Agent Setup](/blog/multi-agent-setup): Configure multiple agents with delegation, scoped tools, and specialized system prompts.
- [How AI agents remember: a survey of persistent memory](/blog/persistent-agent-memory-research): We surveyed how Claude Code, ChatGPT, Cursor, Windsurf, and others implement persistent agent memory — storage formats, compaction, and retrieval.
- [Plans vs tasks: how AI agents think before they act](/blog/plans-vs-tasks-agent-design): A survey of plan-then-execute patterns across Claude Code, Cursor, Devin, Windsurf, and Copilot — and what it means for autonomous agent design.
- [Set Up Gstack](/blog/setup-gstack): Install garrytan's AI engineering workflow agent — one command, batteries-included stack.
- [Set Up Playwright](/blog/setup-playwright): Give your agent full browser automation — headless Chromium, page navigation, form filling via Playwright.
- [Set Up Search](/blog/setup-search): Give your agent web search — zero API keys, one command, instant results.
- [Set Up Telegram](/blog/setup-telegram): Connect a Telegram bot to your CrabTalk agent — one command, real-time responses.
- [Set Up WeChat](/blog/setup-wechat): Connect a WeChat bot to your CrabTalk agent — real-time messaging through WeChat.
- [SOUL.md: brilliant idea, brittle implementation](/blog/soul-md-adoption): We analyzed OpenClaw's SOUL.md personality system — what it gets right about agent identity, and why static files break under production pressure.
- [How AI frameworks control model thinking](/blog/thinking-control): We surveyed seven agent frameworks to understand who controls reasoning depth — the framework, the API, or nobody. The answers split three ways.
- [Tool permissions and the bash bypass problem](/blog/tool-call-permissions): If your agent has bash, do file-editing tools even matter for security? We surveyed eight frameworks and found the answer is architectural, not policy.
- [Use Browser Tools](/blog/use-browser-tools): Give your agent a real browser — navigate pages, extract content, fill forms via agent-browser.
- [Why we built CrabTalk](/blog/why-we-built-crabtalk): Agent runtimes ship memory leaks, CPU spins, and privacy regressions because they're built like prototypes. CrabTalk is built like infrastructure.
- [Write a Skill](/blog/write-a-skill): Create a custom skill — Markdown file, YAML frontmatter, available to all your agents.

# Hub

- [crabtalk/outlook](/hub/crabtalk/outlook): Outlook MCP server for email and calendar via Microsoft Graph API
- [crabtalk/search](/hub/crabtalk/search): Meta-search aggregator (Bing, Brave, DuckDuckGo, Mojeek, Wikipedia)
- [crabtalk/telegram](/hub/crabtalk/telegram): Telegram bot gateway for agents
- [crabtalk/wechat](/hub/crabtalk/wechat): WeChat bot gateway for agents
- [figma/figma](/hub/figma/figma): Figma MCP server for reading designs, creating diagrams, and managing Code Connect
- [garrytan/gstack](/hub/garrytan/gstack): garrytan's AI engineering workflow agent powered by crabtalk
- [microsoft/playwright](/hub/microsoft/playwright): Browser automation with Playwright MCP server and CLI tools
- [notion/notion](/hub/notion/notion): Notion MCP server for pages, databases, and blocks
- [shopify/dev-mcp](/hub/shopify/dev-mcp): Shopify Dev MCP server for docs search, API schema exploration, and theme validation
- [vercel/agent-browser](/hub/vercel/agent-browser): Browser automation CLI for AI agents