How to use AI agents on Stremit
How to use AI agents on Stremit
A complete walkthrough of the Marketplace — from browsing agents to running your first analysis, uploading a resume, and exporting your results.
What are agents on Stremit?
Agents are AI-powered tools built on top of Claude that run directly inside Stremit. Unlike a generic chatbot where you start from a blank prompt every time, agents are pre-engineered for a specific task. You fill in a few inputs, click run, and get back a structured, deeply reasoned output in seconds.
Think of an agent as a specialist you can consult on demand. The Startup Validator thinks like a YC partner. The Learning Roadmap Generator thinks like a curriculum designer. The AI Career Coach thinks like an executive coach who has reviewed thousands of resumes.
Note: Agents are a Pro feature. You need an active Pro subscription to run agents, fork them, and publish your own. You can browse the marketplace and read agent descriptions without a Pro account.
Every agent on Stremit has three parts:
- A system prompt — the engineering behind the agent's behavior. You can read the full prompt on every detail page.
- Variables — input fields you fill in to customize the output for your specific situation.
- An execution engine — Claude Sonnet, running server-side, processing your inputs and returning the result.
The three agents available today
Stremit currently has three official agents in the marketplace, all built and verified by the Stremit team.
🎯 Startup Validator
Official Agent Pro
Runs your startup idea through a 7-layer validation framework. Returns a problem score, market size reality check, kill shot assumption, a 7-day experiment to test it, and a final verdict on whether to build, pivot, or kill the idea.
🗺️ Learning Roadmap Generator
Official Agent Pro
Takes any topic and generates a complete beginner-to-advanced roadmap with phases, a week-by-week timeline, a visual diagram, curated resources for each phase, and export to PDF or PNG.
💼 AI Career Coach
Official Agent Pro
Seven modes covering the full career lifecycle: resume review, interview prep, LinkedIn optimization, salary negotiation, career pivot, personal branding, and job search strategy. Accepts file uploads including PDF, Word, and image scans.
How to run your first agent
Running an agent takes less than a minute. Here is the exact flow from start to result.
Step 1 — Go to the Marketplace
Click Explore in the top navigation and select Marketplace, or go directly to stremit.io/agents. You will see all available agents. The index page is public — no login required to browse.
Step 2 — Click an agent card
Click "Run Agent" button. If you are not a Pro member, you will be taken to the upgrade page. If you are Pro, you go directly to the agent detail page.
Step 3 — Read the agent description
Every agent detail page shows the full description, the variables it accepts, the system prompt it runs, and community reviews. Read what the agent does before filling in your inputs — the context improves output quality.
Step 4 — Fill in the variable fields
Each agent has 2–5 input fields. These are the variables the agent interpolates into its prompt before sending to Claude. The more specific and detailed your inputs, the more specific and useful the output. Vague input produces vague output.
Step 5 — Click "Run with Claude"
The button becomes active once all required fields are filled. Processing takes 10–30 seconds depending on the agent. Do not close the tab while it runs.
Step 6 — Read and use your result
Results appear directly on the page. For the Career Coach, results render with formatted section headers. For the Learning Roadmap, you get a tabbed interface with phases, diagram, timeline, and insights. For the Startup Validator, you get a structured 7-layer analysis.
Pro tip: The Run button stays disabled until all required fields have content. If the button is not activating, check for empty required fields — they are marked with an asterisk.
Using variables to customize output
Variables are the core mechanic of agents. They replace the generic prompt with your specific context, which is why agent output is dramatically more useful than asking Claude a general question.
Here is an example using the Startup Validator. The same agent with different inputs produces completely different, fully tailored analysis:
| Variable | Vague input | Better input |
|---|---|---|
startup_idea |
An app for productivity | A Slack plugin that automatically writes meeting summaries and action items using AI, sent to participants within 60 seconds of the call ending |
target_customer |
Businesses | Remote-first engineering teams of 10–50 people who have daily standups and weekly planning calls but no dedicated ops person to track follow-ups |
your_background |
I work in tech | 8 years as an engineering manager at two Series B startups, 2,000 newsletter subscribers in the devtools space, previously built a Chrome extension with 4,000 active users |
The quality gap between these two inputs in the final output is significant. The agent with vague inputs gives you generic market analysis. The agent with specific inputs gives you analysis that names your actual competitors, identifies your actual unfair advantage, and designs an experiment specific to your actual situation.
For the Career Coach, variables work the same way — but you also get a mode selector. Pick your mode first, and the placeholder text in the other fields updates automatically to tell you exactly what to paste.
The seven available modes are:
| Mode | What it produces |
|---|---|
| 📄 Resume Review | Full resume audit, rewritten summary, bullet point rewrites, ATS keywords, priority fix list |
| 🎯 Interview Prep | 5 likely questions with answer frameworks, objection handling scripts, closing questions, 48-hour prep plan |
| 💼 LinkedIn Optimization | Headline rewrites, full About section rewrite, keyword strategy, content ideas |
| 💰 Salary Negotiation | Market rate analysis, word-for-word negotiation script, responses to the 3 most common pushbacks |
| 🔄 Career Pivot | Transferable skills map, gap analysis, 90-day pivot plan, back door entry strategy |
| ⭐ Personal Branding | Positioning statement, 3 content pillars, platform strategy, 30-day visibility sprint |
| 🔍 Job Search Strategy | Search audit, target company framework, cold outreach script, weekly schedule |
Uploading files for deeper analysis
The AI Career Coach accepts file uploads as context. Instead of pasting your resume as text, you can upload the actual document and Claude reads it directly.
Supported formats:
Files that Claude reads natively:
- PDF — Claude processes the full layout, formatting, and structure of the file directly
- PNG, JPG, WEBP — useful for scanned resumes or screenshots of LinkedIn profiles
Files where text is extracted first:
- DOCX / DOC — text is pulled from the document and injected into the prompt
- TXT — read directly and used as context
How to upload:
Scroll to the upload zone below the mode selector on the Career Coach detail page. Drag and drop your file or click to browse. When Resume Review mode is active, the upload zone is prominent. In other modes it appears as a compact "attach file" option.
Once uploaded, the file processes in a few seconds and shows a preview of extracted text. When you click Run with Claude, the analysis is based on your actual document — not just the text fields.
Privacy: Files are never stored on Stremit's servers. They are processed entirely in memory, used for the single API call, and discarded immediately. Nothing you upload is saved, logged, or accessible after your session ends.
Exporting your results
The Learning Roadmap Generator produces results that are worth saving. After running the agent, you get two export options at the top of the results panel.
Save as PDF
Generates a multi-page PDF document with all phases, resource links, milestones, common mistakes, and a footer that reads "Generated by Stremit · stremit.io". Good for saving to Notion, printing, or sharing with a mentor.
Save as PNG
Captures the full on-screen result as a high-resolution image at 2x pixel density for retina displays. Good for sharing on LinkedIn, Twitter, or in a community Discord.
The diagram tab also has its own "Download diagram" button that exports just the flowchart as a PNG — useful if you want to embed the learning path diagram in a document or presentation.
Saving Career Coach sessions
The Career Coach has a "Save session" button that stores your result in your browser's local storage. Up to ten sessions are saved automatically, organized by mode and date. You can reload any previous session by clicking it in the Saved Sessions panel below the Run button.
Forking and publishing your own agents
Every agent on Stremit can be forked. Forking creates a copy of the agent under your account that you can modify — change the system prompt, update the variables, adjust the behavior — and publish back to the marketplace.
To fork an agent:
Click the "Fork this Agent" button in the right sidebar of any agent detail page. Your fork appears in your profile and in the marketplace with a "Fork of [original agent name]" label.
To publish your own agent from scratch:
Click "+ Share an Agent" at the top of the marketplace index. You will need a Pro account. The create form asks for:
| Field | What to write |
|---|---|
Name |
Clear and specific. What the agent does, not what it is. "Email Rewriter" not "AI Tool". |
Type |
Prompt (structured output), Workflow (JSON/multi-step), or Agent (autonomous reasoning). |
Tagline |
One sentence. What the user gets, not what the agent does. "Get a complete LinkedIn rewrite in 60 seconds" not "An AI that rewrites LinkedIn profiles". |
Content |
Your system prompt. Use {{variable_name}} syntax for any input you want the user to fill in. The platform auto-detects these and creates form fields. |
On variable syntax: Variables are detected automatically from double curly brace syntax in your content. Write
{{topic}}in your prompt and Stremit creates an input field labeled "topic". Users fill in the field and it replaces the placeholder before the prompt is sent to Claude.
Tagging your agent:
When creating an agent, you can tag up to 5 AI tools, collections, stacks, and posts that relate to it. Tags make your agent more discoverable and show users the broader context of what your agent does and what tools it works alongside.
Frequently asked questions
How long does an agent take to run?
Between 10 and 45 seconds depending on the agent and the length of your inputs. The Startup Validator typically returns in 15–20 seconds. The Learning Roadmap Generator takes 30–45 seconds because it generates significantly more structured output. The Career Coach varies by mode — Resume Review is the most detailed and can take up to 45 seconds.
Which model powers the agents?
All official Stremit agents run on Claude Sonnet. This is the best balance of response quality, reasoning depth, and speed for the tasks these agents handle. Additional models including Claude and GPT-5 are coming for Pro users in a future update — these will unlock image search, video upload analysis, and web search beyond Stremit's database.
Can I run an agent more than once?
Yes. There is no limit on how many times you can run an agent. Each run is independent — the agent does not remember your previous runs. If you want to iterate on a result, adjust your variable inputs and run again.
What happens to my uploaded files?
Files are processed in memory and never stored. They exist only for the duration of the API call to Claude. Nothing you upload is written to disk, saved in the database, or accessible after your session ends. The processing happens server-side over an encrypted connection.
Can free users see the marketplace?
Yes. The marketplace index is fully public — anyone can browse agents, read their descriptions, and see their system prompts. Running agents, forking them, and publishing your own requires a Pro subscription.
How do I get my agent verified as Official?
Official status is granted by the Stremit team to agents that meet quality standards: a clear purpose, a well-engineered system prompt, meaningful variables, and demonstrated usefulness to the community. If you have published an agent you believe meets these standards, reach out via the contact page.
Is there a limit on how many agents I can publish?
No hard limit. Pro members can publish as many agents as they like. The marketplace sorts by fork count and engagement, so quality matters more than quantity.
What's next for Agents
The three official agents are the foundation. Here is what is coming:
- More models — Claude and GPT-5 execution for Pro users, with image upload and live web search
- Agent chaining — run the output of one agent directly into another
- Scheduled agents — run an agent automatically on a schedule (weekly digest, daily brief)
- Community agents — as Pro members publish more agents, the marketplace grows into a library of specialized tools built by and for the Stremit community
The best agents will be the ones the community builds. If you have an idea for an agent that would be genuinely useful — something you would use yourself every week — publish it. Fork an existing one and make it better. That is how the marketplace grows.
Published by the Stremit team · Browse the Marketplace · Upgrade to Pro
No comments yet. Start the discussion!