Discover the Best AI Tools

Smithers defines code-maintenance workflows that turn issues into reviewed and tested changes, with optional wiki updates. Its release candidate requires Linux and Node.js 26.4.0, is installed from source rather than npm, and needs credentials for the configured model.

AIAvatarKit builds conversational avatars using interchangeable speech recognition, language-model and speech synthesis components, with tools and channel adapters. Its default quickstart requires Python 3.11+, an OpenAI key and a reachable VOICEVOX-compatible server; other integrations need their own configuration.

This provider connects the Vercel AI SDK to OpenRouter text and embedding models, with experimental decision-model support. The current release targets AI SDK 7, Node.js 22+ and ESM; it needs OpenRouter access, and the alpha Decisions API can change independently.

Hippo stores agent memories locally in SQLite with Markdown mirrors, lexical search and optional embeddings. It requires Node.js 22.16+; initialization can change agent instructions, install hooks and schedule a daily run, while feedback-based ranking does not establish that a memory is true.

Vercel AI CLI provides composable terminal commands for text, image, video and audio generation and typed evaluations. This Labs experiment requires Node.js 22+ and a gateway or provider key; available operations and charges depend on the selected model service.

kiprio MCP Server connects AI clients to hosted utilities for contact checks, geocoding, DNS, web extraction and text processing. The underlying service has anonymous and registered usage limits; inputs go to the service, and checks should not be treated as guaranteed deliverability or complete data redaction.

swagger-testcase-mcp generates categorized test cases and mock data from Swagger/OpenAPI documents, with validation, version comparison and exports. It requires Node.js 18+ and accessible specifications; generated cases are a testing baseline, not evidence that the API passed them.

squiggles bridges coding agents to language servers for diagnostics, navigation and refactoring through MCP. Each configured language server must be installed separately; formatting writes files, and rename can also apply edits when requested.

Featureflip MCP exposes feature-flag management, targeting and cleanup workflows to AI assistants. It requires a Featureflip personal or service token; webhook administration needs higher privileges, and application-side flag evaluation remains the job of language SDKs.

Mobius Studio MCP lets agents plan and edit WordPress sites using Sweipe or FlatMobile themes through their REST interface. It requires the compatible theme plugin, an activated license and a WordPress application password; AI operations consume the theme service's credits.

SuperGrep gives coding agents indexed keyword search, SQL and cloud-assisted semantic retrieval with source citations. Its documented setup needs Node.js 22+ on macOS or Linux; semantic search and delegated queries require an account and a copy of the index in Infino Cloud.

OpenFiles opens files edited by coding agents in VS Code-compatible editors and relays available diagnostics through supported hooks. It relies on installed language servers and configured integrations; diagnostics coverage and feedback support vary by language and agent.

twoperson exchanges review packets and commit-bound verdicts between coding agents through a shared directory. It validates recorded approvals without a server or API key; this record-level rule does not by itself prevent an agent from running Git or deployment commands outside the workflow.

OrcaCode Review integrates AI review findings and severity-based merge checks into GitHub pull requests, with local-agent skills as another mode. CI review uses an OrcaRouter key and configured model; the local mode relies on your agent and does not itself prove code correctness.

argot learns patterns from Git history and flags deviations in changes, using local statistics and embedded code representations. It needs usable history and supported source files; findings are review leads, not proof of defects or evidence of who wrote the code.

Mantyl creates a project handover record from a repository and coding-agent history, linking claims to evidence and verification status. It needs Node.js 20.12+; full sandbox checks require Docker, and reading Cursor history needs Node.js 22.13+.

AgentBridge relays messages between live Claude Code and Codex sessions for task coordination and cross-review. It requires Bun and both agents; this experimental local bridge is not a security boundary, and its default launch commands bypass agent permission prompts.

Loadout selects reusable context kits for coding agents and renders them into local ignored files at launch. Selection is deterministic, not model-driven; it requires a supported agent and equips one kit at a time rather than merging multiple loadouts.

eve is a filesystem-first framework for durable AI agents, organizing instructions, typed tools, skills, message channels and schedules in conventional directories. Its CLI scaffolds a project and starts an interactive terminal interface with a configured model. It is currently beta, so APIs, documentation and behavior may change.

Shubham510’s unofficial TypeSafe Go SDK uses the standard library for typed Noul, Choice and Score calls, retries and model discovery. Its module requires Go 1.27.1 and a TypeSafe API key for remote inference. This is a separate implementation from other typesafe-go repositories, not a bundled local model.

valksor’s unofficial PHP SDK provides typed Noul, Choice and Score requests to TypeSafe, with retries, model discovery, request metadata and injectable transport. It requires PHP 8.4 with cURL and JSON plus a TypeSafe API key. It is a hosted-service client, distinct from other similarly named PHP SDKs.

TypeSafe API by typesend is an alpha Elixir SDK for typed decision requests, with retries, telemetry, concurrent calls and offline test stubs. It requires Elixir 1.18 or newer and a TypeSafe API key for actual hosted inference. Test stubs simulate responses for application tests and are not model inference.

TinyJevClient is a Rust HTTP client for typed Jev decisions, including Noul, Choice and Score, response validation and request diagnostics. Its repository now resolves to tinyhumansai/tinydecisionmodels. It requires Rust 1.88 and provider credentials; this GPL-3.0-only client does not distribute a local model or decide application policy.

typesafe.zig is a community Zig client for TypeSafe, exposing typed Noul, Choice and Score decisions with compile-time checks and standard-library networking. It requires Zig 0.16 and a TypeSafe API key for hosted inference. Type checking validates interface usage, not the factual correctness of model decisions.