Spec on AI Knowledge Base

Symphony Service Specification

Mon, 01 Jan 0001 00:00:00 +0000

Complete technical specification for implementing Symphony: six architectural layers, orchestration state machine, workspace management, agent runner protocol, polling and scheduling, retry semantics, and observability.

Advanced setup

Mon, 01 Jan 0001 00:00:00 +0000

System requirements, platform-specific installation, version management, and uninstallation for Claude Code.

Agent Specs

Mon, 01 Jan 0001 00:00:00 +0000

An Overview of Cohere's Models

Mon, 01 Jan 0001 00:00:00 +0000

Cohere has a variety of models that cover many different use cases. If you need more customization, you can train a model to tune it to your specific use case.

Application-specific evaluation approaches

Mon, 01 Jan 0001 00:00:00 +0000

Architecture

Mon, 01 Jan 0001 00:00:00 +0000

Aspect Critique

Mon, 01 Jan 0001 00:00:00 +0000

Authorization

Mon, 01 Jan 0001 00:00:00 +0000

The MCP authorization specification defines the OAuth 2.0-based flow that remote MCP servers use to authenticate clients, and it is the formal basis for everything the higher-level registry authentication guide simplifies. Focus on the required OAuth grant types, token scoping rules, and the metadata discovery mechanism that clients use to find a server’s authorization endpoint. A key gotcha is that the spec mandates specific OAuth metadata fields that many generic OAuth libraries do not populate by default, so you may need custom configuration even when using well-known auth frameworks. This is dense specification text best read alongside a working implementation for reference.

Backends

Mon, 01 Jan 0001 00:00:00 +0000

Choose and configure filesystem backends for deep agents. You can specify routes to different backends, implement virtual filesystems, and enforce policies.

Backends

Mon, 01 Jan 0001 00:00:00 +0000

Choose and configure filesystem backends for deep agents. You can specify routes to different backends, implement virtual filesystems, and enforce policies.

Browse Collections

Mon, 01 Jan 0001 00:00:00 +0000

Inspect your Chroma collections with an in-terminal UI.

Cancellation

Mon, 01 Jan 0001 00:00:00 +0000

Completion

Mon, 01 Jan 0001 00:00:00 +0000

Count Assistants

Mon, 01 Jan 0001 00:00:00 +0000

Get the count of assistants matching the specified criteria.

Count Crons

Mon, 01 Jan 0001 00:00:00 +0000

Get the count of crons matching the specified criteria.

Count Threads

Mon, 01 Jan 0001 00:00:00 +0000

Get the count of threads matching the specified criteria.

Crafting Effective Agents

Mon, 01 Jan 0001 00:00:00 +0000

Learn best practices for designing powerful, specialized AI agents that collaborate effectively to solve complex problems.

Create custom subagents

Mon, 01 Jan 0001 00:00:00 +0000

Create and use specialized AI subagents in Claude Code for task-specific workflows and improved context management.

Customize Agents

Mon, 01 Jan 0001 00:00:00 +0000

A comprehensive guide to tailoring agents for specific roles, tasks, and advanced customizations within the CrewAI framework.

Delete Model Checkpoints

Mon, 01 Jan 0001 00:00:00 +0000

Delete specific checkpoints for a model.

deprecated-spec

Mon, 01 Jan 0001 00:00:00 +0000

Elicitation

Mon, 01 Jan 0001 00:00:00 +0000

Evaluation

Mon, 01 Jan 0001 00:00:00 +0000

Guide to evaluating LLMs for specific tasks with metrics, human, and LLM-based methods

EXA Search Web Loader

Mon, 01 Jan 0001 00:00:00 +0000

The ‘EXASearchTool’ is designed to perform a semantic search for a specified query from a text’s content across the internet.

Filtering with Where

Mon, 01 Jan 0001 00:00:00 +0000

Learn how to filter search results using Where expressions and the Key/K class to narrow down your search to specific documents, IDs, or metadata values.

Fireworks Agent: Supervised Fine-Tuning

Mon, 01 Jan 0001 00:00:00 +0000

Run end-to-end SFT with Fireworks Agent — dataset inspection, hyperparameter sweep, evaluator-guided selection, and a deployed winner.

Fork a run

Mon, 01 Jan 0001 00:00:00 +0000

Explore different parameters or models from a specific point in an experiment without impacting the original run.

Generate dense embeddings

Mon, 01 Jan 0001 00:00:00 +0000

Generate dense vector embeddings for the given texts using the specified model. Provide either ‘instructions’ or both ’task’ and ’target’ alongside ’texts’.

Generate sparse embeddings

Mon, 01 Jan 0001 00:00:00 +0000

Generate sparse vector embeddings for the given texts using the specified model. Provide either ‘instructions’ or both ’task’ and ’target’ alongside ’texts’. Set ‘fetch_labels’ to true to include token labels in the response.

Generative UI

Mon, 01 Jan 0001 00:00:00 +0000

Understanding AG-UI’s relationship with generative UI specifications

Get Thread State At Checkpoint

Mon, 01 Jan 0001 00:00:00 +0000

Get state for a thread at a specific checkpoint.

Get Thread State At Checkpoint

Mon, 01 Jan 0001 00:00:00 +0000

Get state for a thread at a specific checkpoint.

Groq

Mon, 01 Jan 0001 00:00:00 +0000

Track and monitor Groq’s ultra-fast LPU™ inference with Weave, capturing model calls, performance metrics, and function chains for high-speed LLM applications using Groq’s specialized hardware acceleration.

Index creation error - missing spec parameter

Mon, 01 Jan 0001 00:00:00 +0000

Key Changes

Mon, 01 Jan 0001 00:00:00 +0000

Ledger & Debugging for RL Rollouts

Mon, 01 Jan 0001 00:00:00 +0000

Inspect snapshot history, reset the ledger, and understand how in-flight requests behave during a weight swap.

Lifecycle

Mon, 01 Jan 0001 00:00:00 +0000

The lifecycle specification defines the three phases every MCP connection goes through: initialization, operation, and shutdown. Focus especially on the initialization handshake where client and server exchange capabilities – getting this wrong is the most common source of connection failures. Note that the protocol version negotiation happens here too, so mismatched versions between client and server will fail fast at this stage rather than producing mysterious errors later.

LLM gateway configuration

Mon, 01 Jan 0001 00:00:00 +0000

Learn how to configure Claude Code to work with LLM gateway solutions. Covers gateway requirements, authentication configuration, model selection, and provider-specific endpoint setup.

Log traces to a specific project

Mon, 01 Jan 0001 00:00:00 +0000

Logging

Mon, 01 Jan 0001 00:00:00 +0000

MCP Get

Mon, 01 Jan 0001 00:00:00 +0000

Implemented according to the Streamable HTTP Transport specification.

MCP Inspector

Mon, 01 Jan 0001 00:00:00 +0000

In-depth guide to using the MCP Inspector for testing and debugging Model Context Protocol servers

MCP Post

Mon, 01 Jan 0001 00:00:00 +0000

Implemented according to the Streamable HTTP Transport specification.

Multi-agent patterns

Mon, 01 Jan 0001 00:00:00 +0000

Architect teams of agents that collaborate, escalate, or specialize by capability.

Multimodal Inputs

Mon, 01 Jan 0001 00:00:00 +0000

Use modality-specific user input parts with typed data/url sources in @ag-ui/core

Multimodal Inputs

Mon, 01 Jan 0001 00:00:00 +0000

Use modality-specific user input parts with typed data/url sources in ag_ui.core

Overview

Mon, 01 Jan 0001 00:00:00 +0000

Overview

Mon, 01 Jan 0001 00:00:00 +0000

Oxylabs Scrapers

Mon, 01 Jan 0001 00:00:00 +0000

Oxylabs Scrapers allow to easily access the information from the respective sources. Please see the list of available sources below:

Pagination

Mon, 01 Jan 0001 00:00:00 +0000

PDF Text Writing Tool

Mon, 01 Jan 0001 00:00:00 +0000

The ‘PDFTextWritingTool’ writes text to specific positions in a PDF, supporting custom fonts.

Ping

Mon, 01 Jan 0001 00:00:00 +0000

Plugins reference

Mon, 01 Jan 0001 00:00:00 +0000

Complete technical reference for Claude Code plugin system, including schemas, CLI commands, and component specifications.

Progress

Mon, 01 Jan 0001 00:00:00 +0000

Prompts

Mon, 01 Jan 0001 00:00:00 +0000

Prompts are the least commonly used of the three MCP primitives, but they fill an important niche: reusable, parameterized templates that help users invoke common workflows. Think of them as saved recipes that combine a specific prompt structure with dynamic arguments. Unlike tools and resources, prompts are user-initiated – the user explicitly selects a prompt from a menu rather than the model discovering it automatically. This makes them ideal for standardizing repetitive interactions like “summarize this codebase” or “review this PR.”

Query threads using the SDK

Mon, 01 Jan 0001 00:00:00 +0000

Programmatically fetch and inspect multi-turn conversation threads from your LangSmith projects.

Rate limits

Mon, 01 Jan 0001 00:00:00 +0000

Rate limits are restrictions that our API imposes on the number of times a user or client can access our services within a specified period of time.

REPL

Mon, 01 Jan 0001 00:00:00 +0000

Use the interactive runner to prototype agents and inspect execution step by step.

Resources

Mon, 01 Jan 0001 00:00:00 +0000

Resources are how MCP servers expose data to the model without requiring the model to “call” anything – think of them as files or documents the application can pull into context. The key distinction from tools is that resources are application-controlled (the host decides when to read them), while tools are model-controlled (the model decides when to invoke them). Pay attention to the URI-based addressing scheme, which lets clients discover and subscribe to resource updates. If your use case is primarily about providing context rather than performing actions, resources are usually the better primitive.

Results

Mon, 01 Jan 0001 00:00:00 +0000

Inspect agent outputs, tool calls, follow-up actions, and metadata returned by the runner.

Roots

Mon, 01 Jan 0001 00:00:00 +0000

Sampling

Mon, 01 Jan 0001 00:00:00 +0000

Sampling is the mechanism that allows an MCP server to request LLM completions through the client, effectively enabling servers to leverage AI capabilities without embedding their own model access. This is one of the most powerful and least intuitive parts of the MCP specification because it inverts the typical client-server relationship. Pay careful attention to the human-in-the-loop requirements, since the spec mandates that clients must obtain user approval before fulfilling sampling requests, which has significant UX implications. If you are building agentic MCP servers that need to reason or generate text, understanding this capability is essential.

Schema Reference

Mon, 01 Jan 0001 00:00:00 +0000

Scrape Element From Website Tool

Mon, 01 Jan 0001 00:00:00 +0000

The ‘ScrapeElementFromWebsiteTool’ enables CrewAI agents to extract specific elements from websites using CSS selectors.

Scrape Website

Mon, 01 Jan 0001 00:00:00 +0000

The ‘ScrapeWebsiteTool’ is designed to extract and read the content of a specified website.

Search runs

Mon, 01 Jan 0001 00:00:00 +0000

Learn how to search for specific runs by name or ID in your project’s Runs table or Workspace.

Selenium Scraper

Mon, 01 Jan 0001 00:00:00 +0000

The ‘SeleniumScrapingTool’ is designed to extract and read the content of a specified website using Selenium.

SEP Guidelines

Mon, 01 Jan 0001 00:00:00 +0000

Specification Enhancement Proposal (SEP) guidelines for proposing changes to the Model Context Protocol

SEP-986: Specify Format for Tool Names

Mon, 01 Jan 0001 00:00:00 +0000

Specify Format for Tool Names

SEP-986: Specify Format for Tool Names

Mon, 01 Jan 0001 00:00:00 +0000

Specify Format for Tool Names

Skills

Mon, 01 Jan 0001 00:00:00 +0000

Use skills to give your agents access to specific capabilities.

Specification

Mon, 01 Jan 0001 00:00:00 +0000

Complete A2A protocol specifications

Specification

Mon, 01 Jan 0001 00:00:00 +0000

Specification Enhancement Proposals (SEPs)

Mon, 01 Jan 0001 00:00:00 +0000

Index of all MCP Specification Enhancement Proposals

Specification Enhancement Proposals (SEPs)

Mon, 01 Jan 0001 00:00:00 +0000

Index of all MCP Specification Enhancement Proposals

Specify a custom run ID

Mon, 01 Jan 0001 00:00:00 +0000

How to specify a custom run ID when tracing with LangSmith.

Speculative Decoding

Mon, 01 Jan 0001 00:00:00 +0000

Speed up generation with draft models and n-gram speculation

Spider Scraper

Mon, 01 Jan 0001 00:00:00 +0000

The ‘SpiderTool’ is designed to extract and read the content of a specified website using Spider.

Structured model outputs

Mon, 01 Jan 0001 00:00:00 +0000

Understand how to ensure model responses follow specific JSON Schema you define.

Subagent streaming

Mon, 01 Jan 0001 00:00:00 +0000

Display specialist subagents with streaming content, progress tracking, and collapsible cards

Subagent streaming

Mon, 01 Jan 0001 00:00:00 +0000

Display specialist subagents with streaming content, progress tracking, and collapsible cards

Tasks

Mon, 01 Jan 0001 00:00:00 +0000

Terminate Session

Mon, 01 Jan 0001 00:00:00 +0000

Implemented according to the Streamable HTTP Transport specification.

Text & Vision Fine-tuning

Mon, 01 Jan 0001 00:00:00 +0000

Fine-tune Mistral’s text and vision models with custom datasets in JSONL format for domain-specific or conversational improvements

Time travel

Mon, 01 Jan 0001 00:00:00 +0000

Inspect, navigate, and resume from any checkpoint in the conversation history

Time travel

Mon, 01 Jan 0001 00:00:00 +0000

Inspect, navigate, and resume from any checkpoint in the conversation history

Tools

Mon, 01 Jan 0001 00:00:00 +0000

Tools are the most commonly used MCP primitive – they let the model invoke server-side functions with structured inputs and receive results. Focus on the JSON Schema-based input validation, which is how the model knows what arguments a tool accepts. A critical detail is that tool calls are model-initiated but require human approval in most host implementations, so design your tool descriptions to be clear enough that users understand what they are approving. Keep tool names concise and descriptions precise, as the model relies heavily on them for deciding when and how to call each tool.

Transports

Mon, 01 Jan 0001 00:00:00 +0000

MCP defines two transport mechanisms: stdio (standard input/output) and HTTP with Server-Sent Events (SSE). For local development and CLI-based servers, stdio is simpler and the right default choice. HTTP+SSE is needed when the server runs remotely or must handle multiple clients. A common gotcha is that the Streamable HTTP transport replaced the older SSE-only transport in the 2025-11-25 spec revision, so be careful not to follow outdated examples that use the previous approach.

Triggers Overview

Mon, 01 Jan 0001 00:00:00 +0000

Understand how CrewAI AMP triggers work, how to manage them, and where to find integration-specific playbooks

Versioning

Mon, 01 Jan 0001 00:00:00 +0000

View a specific run in a project

Mon, 01 Jan 0001 00:00:00 +0000

Learn how to view a specific logged run and its properties using the W&B App or the LEET terminal UI.

Websearch

Mon, 01 Jan 0001 00:00:00 +0000

Websearch enables models to browse the web for real-time, up-to-date information and access specific websites