The AI tools guide
Find tools for coding, agents, research, automation, writing, images, video and audio. Compare the job, access and limits, then use one of six practical guides or explore a sourced product profile.
CodexOpenAIExplore unfamiliar code, work out how to check a change and make the edit with help from OpenAI’s terminal coding assistant.
The directory
Higgsfield puts image and video generation, reference-based editing and preset-led creative work in one browser platform. It combines its own model families, including Soul, with separately identified partner models such as Kling and Veo; choosing the platform does not choose a single generation engine.
Illustrative workflowA concept frame and one camera move for a synthetic training videoRunway offers direct image/video tools, task-specific Apps, a conversational creative Agent and node-based Workflows. These are different ways to organize a production job rather than interchangeable model names.
Illustrative workflowA single motion shot from an owned illustrationFirefly is both Adobe’s generative model family and an application that also offers partner models. An asset made in the application therefore needs a model record before Adobe-model training claims can be attached to it.
Illustrative workflowA background variation for a synthetic workshop announcementMidjourney generates and edits images and can animate a starting image into video. It is suited to exploring a visual direction before committing to a finished asset: composition, lighting and style can be compared against one brief, then a selected draft can be refined.
Illustrative workflowThree visual directions for a synthetic reliability workshopCanva AI works inside Canva’s Visual Suite to turn a brief into a design that can be edited further. Its Design Model announcement describes layered output, while the live product/help pages retain account-dependent rollout and allowances; not every account has identical AI 2.0 features.
Illustrative workflowA workshop announcement with editable copy and layoutMake the choice around your work
Start with a task above, or add up to three tools. These starting sets are examples, not rankings.
Three ways to read the landscape
These maps explain the directory’s organization. Category membership is editorial task fit, not a measured ranking or a complete product inventory.
A tool can belong to several jobs. Follow a category to filter the same directory above.
The company builds or offers the named surfaces below. Applications, models and developer platforms occupy different layers; an account for one does not establish access to another.
An illustrative workflow for choosing and evaluating a tool, not an executed benchmark.
A code patch, cited answer, workflow, image, video or audio deliverable.
Compare the product profile’s account, platform, input rights and processing path.
Use the proposed public or synthetic inputs. Read a practical guide when one is available.
Inspect the citations, behavior, rights and error paths before relying on the result.
02Learning paths
Make a code change, review a pull request, or coordinate a few agents. These walkthroughs show how to use the tools along the way.
Learning path · 5 steps
Start with the supplied support-service client and finish with a reviewed patch, regression tests and a short explanation of its behavior. The same two files carry the work from the first question to the final check: one mistakes an HTTP error for ticket data, while the other repeats a write after an ambiguous failure.
Leave withYou will produce two corrected modules, focused regression tests, and a change note a reviewer can use. No web server, real tickets or package installation is needed for the JavaScript project. Install and authorize one of the linked assistants separately.
Unpack the starter into a disposable folder and install one assistant you are authorized to use. Read the README and run the happy-path tests. Run each contract check separately to see the supplied defects before asking for a fix. These failures are expected in the teaching project.
CheckpointBoth baseline tests pass. The 503 contract and ambiguous-write contract fail for reasons you can explain.
Read the Codex guideThese deeper practical walkthroughs are planned. A sourced product profile may already be available above.
How much model can your hardware comfortably run, and what work will it be good for? This planned guide will connect model choice and local processing with a task you can try. It is not ready yet.
A planned guide to giving an assistant a small approved source to search, then following its answers back to the original material. Setup and the worked example are still being reviewed.
Which tool leaves you with more useful work after mistakes, retries and review time are counted? This planned guide will explain how to compare the same tasks fairly. It is not ready yet.
A planned guide to following an AI request through its model and tool calls, so you can understand a failure or an unexpected bill. The guide is still being prepared.
For building an application that calls OpenAI models directly. The planned guide will cover access, authentication, usage controls and data handling. If you want help working in a repository, the Codex guide is available now.
For applications that call Anthropic models directly. Access, authentication, usage controls and data handling will be covered in a separate guide. For work inside a codebase, start with Claude Code.
Planned coverage will connect hardware requirements and model choice with a useful local task. The installation instructions and example still need a full documentation review.
The future guide will explain setup, model choice and where your data is processed. Its first worked example is still to be reviewed.
A planned guide to understanding which provider handles a model request, what terms apply and how it is billed. The profile is not ready yet.
The planned guide will follow an automation from its input through the services and credentials it needs. Setup and the worked example still need review.
Planned coverage will explain how to inspect model requests and evaluate their results, including what information is sent to the service. The guide is still being prepared.
A planned guide to examining AI behavior and tracing where evaluation data goes. Setup and the worked example still need a full review.
The planned guide will show how to compare results on the same task and understand what the tests establish. Current installation instructions and the example still need review.
A planned guide to how an agent workflow tracks its progress and resumes after an interruption. The setup and first worked example still need review.
