Wispr Flow: Dictation That Works EverywhereStop typing. Start speaking. 4x faster.
Coryant
Coordinated research agents. Running on your Claude.
15 followers
Coordinated research agents. Running on your Claude.
15 followers
Coryant is a set of Claude agents packs for people who don't touch a terminal. Ask a question and 10 coordinated research agents pull market, competitor, pricing, and signal data then hand you one sourced verdict instead of ten browser tabs. GTM, product, career, CA and pack for non technical AI users, built entirely for Claude code and chat interface. No CLI, no GitHub, no setup. Just describe what you need to know.
how do the 10 agents decide when to dig deeper versus just returning the first solid source they find
@kaderhasansoev Hey Kader,
Each agent has a source priority stack baked into its mandate primary sources (regulatory bodies, primary research, official filings) always get exhausted before secondary ones. But the "dig deeper" decision is handled by the Coordinator, not the individual agents. After all agents return findings, the Coordinator cross references them and flags any claim supported by only one source as unconfirmed which triggers a second pass search on that specific claim before synthesis. So it's not the agent deciding to go deeper, it's the pipeline architecture that forces verification before anything reaches the final report. The Adversarial Review layer then stress-tests the top conclusions with disconfirming searches actively trying to break the findings before they're handed to you. That's why claims in the output are graded A/B/C you can see exactly how many sources back each one.
How does it actually pull the signal data behind the verdict, like is it scraping live sources or pulling from a fixed knowledge base that you update on your end?
@ferhat1ffx Hey Ferhat,
Live sources, not a fixed knowledge base and nothing on our end to maintain.
Every run triggers real web searches inside your own Claude session. The agents don't pull from a Coryant database they search the live web in real time, which means the research is current to the day you run it, not to whenever we last updated something on our side.
The architecture is deliberately stateless on our end. We ship the pipeline the agent mandates, source priority stacks, coordination logic, and report structure. Claude's web search does the actual retrieval. So when the GTM pack's Competitor Agent looks for G2 reviews or funding signals, it's searching right now, not returning cached results.
The practical implication: run the same brief two weeks apart and you'll get different findings if the market moved. That's a feature, not a bug especially for signal based decisions where a 30 day old competitor analysis is already stale.
Tried the career pack and it actually pulled salary ranges and interview themes I hadn't seen elsewhere, all cited. Wish I'd had this before my last job search.
@saniyekayoglsp Hey Saniye,
"Wish I'd had this before my last job search" is going to sit with me that's the exact feeling the Career pack was built around. The salary and interview theme research comes from the Consumer Agent cross referencing community discussions, Glassdoor patterns, and role specific forums that don't usually surface in a single search. The citations are non-negotiable for us an uncited salary range is just a guess dressed up as research.
If you're in an active search now or know someone who is, the pack hits differently when you're running it on a specific company before a real interview. The output changes the conversation you walk into.
Tried the GTM pack on a small launch and the sourced verdict actually saved me a couple hours of tab juggling, which I did not expect. Clean answer, no setup, felt like having a junior analyst on call.
@lyasczpu Hey Ilyas,
This is exactly the moment the pack was built for the pre launch tab juggling that eats the exact hours you need for execution. "Junior analyst on call" is the most accurate description anyone's given it, honestly. Glad the sourced verdict landed cleanly. What was the launch for? Curious what the Competitor or Signals agent surfaced that was most useful.
Asked it to map the competitive landscape for a niche product idea and the verdict came back with sources I actually wanted to click through, way more useful than a single chat reply. The no-setup angle is real, my non-technical coworker got it running in chat without me hovering.
@salimtemencska Hey Salim,
The "sources I actually wanted to click through" part is the whole bet if the report just tells you the answer without showing you where it came from, you can't trust it enough to act on it. Evidence grading exists exactly for that reason.
And the non technical coworker getting it running without you hovering that's the thing we obsess over. Most agent pipelines assume you're comfortable in a terminal. Coryant assumes you're comfortable in Claude's chat window and nothing else.
What was the niche? Always curious what categories the Competitor Agent has to work hardest in.
How do the 10 agents stay coordinated without overlapping each other's research, and does the final verdict actually cite which agent pulled what or does it just feel like one smooth answer?