AI Tools
AI tools are best thought of as “capabilities” you can mix and match: writing, coding, images, audio, video, research, automation, and privacy-first local models. This page helps you pick the right category quickly, avoid common mistakes (like hallucinations), and start with prompts you can copy and adapt.
Pick a category that matches your output (text, code, image, audio, video) and your risk level (public vs private data). Use a simple workflow: draft → critique → refine → verify. If accuracy matters, require sources and double-check key claims before sharing.
All tools
Start broad, then narrow with tabs + search.
Chat & writing assistants
Chat & writingThese tools generate and rewrite text: emails, docs, plans, marketing copy, and explanations. They shine when you can give clear context (audience, tone, constraints) and when you iterate: ask for a draft, then ask for improvements. They are not a source of truth—treat them like a fast collaborator, not a fact database.
Coding assistants
CodingCoding assistants help you write, explain, refactor, and debug code. They’re strongest when you provide the exact context: tech stack, file structure, error messages, and expected behavior. They can speed you up, but you still need to review logic, security, and edge cases—especially for auth, payments, and data handling.
Image generation
ImageImage tools generate new visuals from text prompts or transform existing images (style, background removal, variations). They’re great for quick concepts, placeholders, and mood exploration before you spend time on final design. For production work, you still need to check licensing, brand rules, and whether the image is actually usable at your required resolution.
Audio (speech-to-text, text-to-speech)
AudioAudio tools turn speech into text (transcription) and text into speech (voice). They’re useful for meetings, interviews, podcasts, accessibility, and voice interfaces. The biggest risks are privacy (audio can include sensitive info) and accuracy (names, numbers, and domain terms often need correction).
Video (summaries, captions, basic generation concepts)
VideoVideo tools help you turn long videos into summaries, chapters, highlights, and captions—and some can generate short clips from prompts or storyboards. They shine for editing workflows where you need structure and reuse (clips for social, searchable archives). Expect to review outputs: timestamps, names, and “what was actually said” are easy to get wrong.
Research & knowledge tools (search, retrieval, citation workflows)
ResearchResearch tools help you find, organize, and cite information—often by combining web search with retrieval from your documents (RAG). They’re best when you care about correctness and want traceable sources. A strong research workflow is: search → read → extract → cite → write, rather than “ask once and trust the answer.”
Automation & “agent” style tools (workflows, integrations)
Automation & agentsAutomation and agent-style tools connect steps together: trigger → gather context → take actions → report results. They’re great for repetitive work (ticket triage, daily reports, content pipelines) and for coordinating multiple tools. The key is guardrails: clear permissions, limited actions, and logs you can audit.
On-device / local models (privacy-first workflows)
On-device / localLocal models run on your device or your own server, keeping data closer to you. They’re a great fit when privacy, compliance, or offline access matters. The tradeoff is setup effort and sometimes lower quality or smaller context windows compared to top cloud models.
Chat & writing assistants
These tools generate and rewrite text: emails, docs, plans, marketing copy, and explanations. They shine when you can give clear context (audience, tone, constraints) and when you iterate: ask for a draft, then ask for improvements. They are not a source of truth—treat them like a fast collaborator, not a fact database.
Great for
- Drafting emails, proposals, and internal docs faster
- Turning rough notes into a clean structure
- Rewriting for tone (friendly, formal, concise) and clarity
- Summarizing long text into action items and next steps
- Brainstorming outlines, names, and messaging angles
Limitations / gotchas
- Hallucinations: it may invent details, quotes, or references—verify anything factual
- Privacy risk: pasted content might be stored or used for training depending on the provider
- Style drift: it can start helpful then become generic unless you keep constraints explicit
- Copyright and licensing: avoid asking it to reproduce copyrighted text or imitate a living author’s style too closely
How to choose
- ✓Pick a tool that supports long context if you paste lots of background
- ✓Prefer tools with editing modes (rewrite/shorten/expand) if you iterate often
- ✓If accuracy matters, choose tools that support citations or browsing with sources
- ✓Check data handling settings (retention, training opt-out) before pasting sensitive info
- ✓Decide whether you need “creative” output or “grounded” output and tune settings accordingly
Recommended workflow
- 1State your goal + audience + tone in one sentence
- 2Ask for an outline before a full draft
- 3Request a critique: confusing parts, missing info, risks
- 4Revise with constraints (length, must-include points, forbidden claims)
- 5Fact-check key claims and add links/citations where needed
Prompt templates
You are a writing assistant. Goal: [goal] Audience: [audience] Tone: [tone] Constraints: [constraints] Must include: [mustInclude] Avoid: [avoid] Write a first draft. Then list 5 questions you would ask me to improve accuracy or completeness.
Tip: Replace placeholders like [topic] and keep any sensitive info anonymized.
Rewrite the text below for [audience] in a [tone] tone. Constraints: - Keep meaning the same (do not add new facts) - Keep it under [length] - Use short sentences and clear headings Text: [paste text here]
Tip: Replace placeholders like [topic] and keep any sensitive info anonymized.
Coding assistants
Coding assistants help you write, explain, refactor, and debug code. They’re strongest when you provide the exact context: tech stack, file structure, error messages, and expected behavior. They can speed you up, but you still need to review logic, security, and edge cases—especially for auth, payments, and data handling.
Great for
- Generating boilerplate (components, routes, tests, types) quickly
- Debugging with error logs and reproduction steps
- Refactoring for readability and performance
- Explaining unfamiliar code and suggesting safer patterns
- Writing unit tests and describing edge cases
Limitations / gotchas
- Incorrect assumptions: it may guess your project structure or dependencies
- Security gotchas: it can suggest unsafe auth, weak validation, or insecure storage
- Outdated APIs: examples may not match your framework version
- Licensing and code provenance: avoid pasting proprietary code into tools that can retain or train on it
How to choose
- ✓Prefer tools that can reference your repo or a selected set of files (with permission)
- ✓Pick assistants that understand your stack (Next.js, TypeScript, Tailwind, etc.)
- ✓Look for tools with test generation and code review modes
- ✓If you work with secrets, choose tools that support redaction and enterprise controls
- ✓Make sure the tool can run or validate code (lint/test) or integrates with your CI
Recommended workflow
- 1Describe the bug and expected behavior; include minimal reproduction steps
- 2Paste the exact error message + relevant code snippet (not the whole app)
- 3Ask for 2–3 hypotheses and how to verify each
- 4Apply the fix and run lint/tests; request test cases for regression
- 5Do a quick security review (input validation, authz, secrets, logging)
Prompt templates
You are a senior engineer. Stack: [stack] Problem: [what’s broken] Expected: [expected behavior] Actual: [actual behavior] Error/logs: [paste logs] Relevant code: [paste code] Give 3 likely root causes, how to confirm each, and the safest fix. Include any edge cases and tests I should add.
Tip: Replace placeholders like [topic] and keep any sensitive info anonymized.
Refactor the following code for [goal: readability/perf/maintainability]. Constraints: - No behavior changes - Keep public function signatures the same - Add types where missing - Explain each change briefly Code: [paste code]
Tip: Replace placeholders like [topic] and keep any sensitive info anonymized.
Image generation
Image tools generate new visuals from text prompts or transform existing images (style, background removal, variations). They’re great for quick concepts, placeholders, and mood exploration before you spend time on final design. For production work, you still need to check licensing, brand rules, and whether the image is actually usable at your required resolution.
Great for
- Concept art and rapid visual exploration
- UI illustration directions (style frames, icon directions, hero imagery ideas)
- Product mockups and background variations
- Marketing assets for early drafts (then replace with approved visuals)
- Generating multiple options quickly (colors, composition, lighting)
Limitations / gotchas
- Licensing and rights can be unclear—confirm usage rights for commercial work
- Inconsistency: characters, logos, and details can change across generations
- Text in images is often wrong or garbled; don’t rely on it for UI copy
- Privacy risk if you upload user photos or sensitive documents—assume it may be stored
How to choose
- ✓If you need brand-safe output, prefer tools with style control and reference images
- ✓Pick tools that support consistent output (seeds, image-to-image, style presets)
- ✓Check resolution options and upscaling for your target use (web, print)
- ✓Confirm licensing terms for your use case (commercial, internal, editorial)
- ✓If you upload images, review retention policies and opt-out settings
Recommended workflow
- 1Write a short style brief (mood, palette, composition, do/don’t)
- 2Generate 6–12 variants and pick 1–2 directions
- 3Iterate with targeted changes (lighting, framing, background, style)
- 4Run a rights/brand check and replace any risky elements
- 5Export in the correct size and compress for delivery
Prompt templates
Create an image in this style: - Subject: [subject] - Setting: [setting] - Mood: [mood] - Color palette: [palette] - Composition: [composition] - Lighting: [lighting] - Avoid: [avoid] - Aspect ratio: [ratio] Generate 4 variations with small differences in composition.
Tip: Replace placeholders like [topic] and keep any sensitive info anonymized.
Generate a clean, modern hero illustration for a website about [topic]. Constraints: - Minimal, friendly, lots of whitespace - Works behind a headline and 2 buttons - No text inside the image - Style: [flat/3D/line art], colors: [colors] - Aspect ratio: [ratio] Output: 6 options with different compositions.
Tip: Replace placeholders like [topic] and keep any sensitive info anonymized.
Audio (speech-to-text, text-to-speech)
Audio tools turn speech into text (transcription) and text into speech (voice). They’re useful for meetings, interviews, podcasts, accessibility, and voice interfaces. The biggest risks are privacy (audio can include sensitive info) and accuracy (names, numbers, and domain terms often need correction).
Great for
- Transcribing meetings and interviews into searchable notes
- Creating voiceovers for demos, tutorials, and accessibility
- Adding captions and improving content discoverability
- Voice UX prototypes (scripts, timing, tone) without studio work
- Summarizing long recordings into key decisions and action items
Limitations / gotchas
- Accuracy drops with accents, overlapping speakers, noise, and jargon
- Privacy: recordings may contain sensitive data; treat audio like confidential documents
- Consent and compliance: you may need permission to record or process voices
- Synthetic voices can raise rights/consent issues—avoid impersonation
How to choose
- ✓If you need high accuracy, choose tools with diarization (speaker labels) and domain vocab
- ✓For voiceovers, look for control over pace, emotion, and pronunciation
- ✓Check export formats (SRT/VTT for captions, WAV/MP3 for audio)
- ✓Confirm data retention and whether audio is used for training
- ✓If compliance matters, pick tools with enterprise controls and regional hosting
Recommended workflow
- 1Record clean audio (close mic, reduce noise, avoid crosstalk)
- 2Transcribe with speaker labels; spot-check names and numbers
- 3Summarize into decisions, action items, and open questions
- 4Generate captions (SRT/VTT) and review timing quickly
- 5Store audio/transcripts securely and redact sensitive sections
Prompt templates
Summarize the transcript below. Output format: - Decisions (bullet list) - Action items (owner + due date placeholders) - Risks / open questions - Key quotes (only if present; do not invent) Transcript: [paste transcript here]
Tip: Replace placeholders like [topic] and keep any sensitive info anonymized.
Write a voiceover script for a [length]-minute video. Topic: [topic] Audience: [audience] Tone: [tone] Constraints: short sentences, easy to read aloud, avoid jargon. CTA: [call to action] Return: 1) Script 2) A list of words that might be hard to pronounce
Tip: Replace placeholders like [topic] and keep any sensitive info anonymized.
Video (summaries, captions, basic generation concepts)
Video tools help you turn long videos into summaries, chapters, highlights, and captions—and some can generate short clips from prompts or storyboards. They shine for editing workflows where you need structure and reuse (clips for social, searchable archives). Expect to review outputs: timestamps, names, and “what was actually said” are easy to get wrong.
Great for
- Auto-generating captions and translating subtitles
- Creating chapters, highlights, and short social clips
- Summarizing webinars and lectures into notes
- Extracting quotes, key moments, and action items
- Drafting a storyboard or shot list for simple videos
Limitations / gotchas
- Caption timing and speaker attribution often need manual review
- Summaries can miss nuance or overconfidently “interpret” intent
- Copyright: don’t upload or republish content you don’t have rights to
- Generation quality varies; brand-safe, high-fidelity video is still hard
How to choose
- ✓If you need captions, prefer tools that export SRT/VTT and support translation
- ✓For highlights, look for chaptering, keyword search, and clip extraction
- ✓Check supported input formats and max file size/duration
- ✓Understand rights: who owns outputs and what you can publish
- ✓If accuracy matters, choose tools that provide timestamps and source references
Recommended workflow
- 1Upload or link the video; ensure you have usage rights
- 2Generate transcript + captions; spot-check names and numbers
- 3Create chapters/highlights; verify timestamps quickly
- 4Export clips/captions; do a final watch for correctness
- 5Publish with proper attribution and rights notes if needed
Prompt templates
Using the transcript below, create chapters. Requirements: - 6–12 chapters - Each chapter: timestamp placeholder [mm:ss], title, 1-sentence summary - Keep titles short and clear Transcript: [paste transcript here]
Tip: Replace placeholders like [topic] and keep any sensitive info anonymized.
I have a [length]-minute video about [topic]. Goal: create [N] short clips for [platform]. Audience: [audience] Tone: [tone] Constraints: each clip 20–45 seconds, hook in first 2 seconds. Based on this transcript, propose clip segments with: - Start/end timestamp placeholders - Hook line - On-screen text suggestion (short) Transcript: [paste transcript here]
Tip: Replace placeholders like [topic] and keep any sensitive info anonymized.
Research & knowledge tools (search, retrieval, citation workflows)
Research tools help you find, organize, and cite information—often by combining web search with retrieval from your documents (RAG). They’re best when you care about correctness and want traceable sources. A strong research workflow is: search → read → extract → cite → write, rather than “ask once and trust the answer.”
Great for
- Finding credible sources and building a bibliography quickly
- Answering questions grounded in your docs (policies, manuals, notes)
- Creating summaries with citations you can verify
- Comparing claims across sources and spotting disagreements
- Building internal knowledge bases for teams
Limitations / gotchas
- Hallucinated citations can happen—always open links and confirm they support the claim
- Search bias: results depend on queries and ranking; you may miss key sources
- Privacy: uploading documents may be risky without strong controls
- Stale info: models can repeat outdated facts if you don’t force fresh sources
How to choose
- ✓If you need accuracy, choose tools that show sources next to claims
- ✓For internal docs, prefer strong retrieval controls (collections, permissions, re-indexing)
- ✓Look for export options (notes, citations, links) to keep your work reusable
- ✓Decide whether you need web browsing, private docs, or both
- ✓Check how the tool handles citation formatting (URLs, footnotes, quotes)
Recommended workflow
- 1Define the research question and success criteria
- 2Collect sources (web + internal docs) and skim for relevance
- 3Extract key facts with citations beside each fact
- 4Write the output; mark uncertain areas explicitly
- 5Do a final verification pass: open sources and confirm each important claim
Prompt templates
Research question: [question] Requirements: - Only include claims that are directly supported by sources - For each claim, include: claim → source link → 1–2 supporting quotes or notes - If sources disagree, show both sides - If you’re unsure, say “uncertain” Sources I can use: - Web search - Documents: [doc list] Start by listing the top 8–12 sources you would consult and why.
Tip: Replace placeholders like [topic] and keep any sensitive info anonymized.
Write a [format: memo/blog/report] using ONLY the notes below. Constraints: - Do not add new facts - Keep it under [length] - Add a citations section with the provided URLs Notes: [paste notes with URLs here]
Tip: Replace placeholders like [topic] and keep any sensitive info anonymized.
Automation & “agent” style tools (workflows, integrations)
Automation and agent-style tools connect steps together: trigger → gather context → take actions → report results. They’re great for repetitive work (ticket triage, daily reports, content pipelines) and for coordinating multiple tools. The key is guardrails: clear permissions, limited actions, and logs you can audit.
Great for
- Recurring workflows (daily summaries, monitoring, report generation)
- Integrations across apps (docs, email, Slack, CRM, databases)
- Batch tasks (tagging, routing, formatting, data cleanup)
- Semi-automated research pipelines with human review
- Prototype “AI coworkers” that follow checklists
Limitations / gotchas
- Action risk: the tool can delete, email, or publish the wrong thing without safeguards
- Tool failures: APIs change, permissions expire, and edge cases break flows
- Privacy and access: connecting many systems increases blast radius
- Over-automation: processes can become hard to debug if you skip logging and approvals
How to choose
- ✓Start with read-only actions; add write actions only with approvals
- ✓Pick tools with logs, versioning, and easy rollback
- ✓Prefer granular permissions (per integration, per workspace, per dataset)
- ✓If reliability matters, choose retries, rate limits, and alerting
- ✓Decide if you need a no-code builder or code-based orchestration
Recommended workflow
- 1Define the trigger and a clear success metric
- 2Map steps and required permissions; add an approval checkpoint
- 3Run in a sandbox with test data and verbose logging
- 4Launch with read-only mode or “dry run” reports
- 5Monitor results and iterate on guardrails and prompts
Prompt templates
Design an automation workflow for [process]. Inputs: [inputs] Systems: [systems/integrations] Constraints: [constraints] Risks: [risks] Return: 1) Step-by-step flow (trigger → actions → outputs) 2) Required permissions 3) Failure modes + mitigations 4) Human approval points 5) Logging/metrics checklist
Tip: Replace placeholders like [topic] and keep any sensitive info anonymized.
You are an agent that helps with [job]. Rules: - Ask clarifying questions before taking actions - Never run destructive actions without confirmation - Summarize what you will do before doing it - Keep a log of decisions Task: [task] Context: [context] Constraints: [constraints] First, propose a plan and list what you need from me.
Tip: Replace placeholders like [topic] and keep any sensitive info anonymized.
On-device / local models (privacy-first workflows)
Local models run on your device or your own server, keeping data closer to you. They’re a great fit when privacy, compliance, or offline access matters. The tradeoff is setup effort and sometimes lower quality or smaller context windows compared to top cloud models.
Great for
- Privacy-sensitive drafting and analysis (internal docs, client data)
- Offline workflows (travel, restricted networks, field work)
- Custom tooling and integrations inside your own environment
- Cost control for high-volume tasks
- Experimentation with fine-tuning or domain adapters (where appropriate)
Limitations / gotchas
- Hardware limits: speed and quality depend on your CPU/GPU and memory
- Tooling complexity: installation, updates, model management, and monitoring
- Smaller context windows or weaker reasoning in some models
- Licensing varies: some models restrict commercial use or redistribution—read terms carefully
How to choose
- ✓Start with your constraints: privacy needs, offline requirement, hardware budget
- ✓Pick a model size you can run comfortably with your latency target
- ✓Check the model license for your use (commercial/internal/distribution)
- ✓Decide whether you need embeddings/RAG locally as well
- ✓Prefer setups with easy updates, model switching, and audit logs
Recommended workflow
- 1Decide what must stay local (inputs, outputs, logs, embeddings)
- 2Set up the runtime and test with non-sensitive sample data
- 3Create a prompt template library and a lightweight evaluation checklist
- 4Add retrieval (RAG) for your local docs if accuracy matters
- 5Review outputs and tune prompts before using with real data
Prompt templates
You are a local assistant. Treat all input as confidential. Task: [task] Context: [context] Constraints: [constraints] Do not include any sensitive values in the final output. If you need to reference them, use placeholders like [CLIENT_NAME] or [ACCOUNT_ID].
Tip: Replace placeholders like [topic] and keep any sensitive info anonymized.
Answer using ONLY the provided documents. Question: [question] Documents available: [doc collection name] Requirements: - If the docs don’t contain the answer, say “not found in documents” - Quote or reference the exact section you used - Keep the answer under [length]
Tip: Replace placeholders like [topic] and keep any sensitive info anonymized.
Picking the right tool quickly
Safety & privacy
- Don’t paste secrets (API keys, passwords, tokens) into prompts. Rotate any secret you accidentally shared.
- Anonymize sensitive data: remove names, emails, account IDs, and replace with placeholders before sending.
- Verify facts and numbers. Treat model output as a draft until you confirm it from reliable sources.
- Cite sources for factual claims when publishing or making decisions—especially for research and comparisons.
- Respect copyright and licensing: don’t request copyrighted text verbatim, and confirm rights for generated media.
- Be careful with medical/legal/financial advice: use AI for questions and summaries, not as a final authority.
- Assume inputs may be logged. Read retention/training settings and prefer privacy-first tools for sensitive work.
- Use least privilege: in automations, start with read-only permissions and require approval for destructive actions.