Try it in Claude Code:
claude plugin marketplace add relevan-dev/agent-plugins
claude plugin install relevan@relevan
Then /mcp -> relevan -> Authenticate
We can't post as fast as we're shipping 🚢
Live this week:
- Hosted MCP server
- Claude Code plugin
- Codex plugin
- ChatGPT connector
- pip install relevan
- $20 Hobby plan
Search over your data, built for agents.
https://t.co/Nzisx0TQbP
87 PRs opened, 44 merged, 130k lines of code. Last 24 hours.
Speed is dangerous, so we tuned our workflow to make sure everything we ship:
- matches its original intent
- is secure
- won't take down prod 🙏
How are you shipping faster?
https://t.co/ZDDHgUnse8
'Bulb' returned 0 results on a Pokémon card catalog. Bulbasaur was right there. Unless you're monitoring, you'd never know.
Relevan caught 27 zero-result queries in 250 searches. Prefix + phonetic matching: 7. More tweaks: 5, none with a real match.
https://t.co/oCAE0LOnqE
@RelevanDev got accepted into AWS Accelerate 🎉
We're building search that learns: every index tuned on its own data, every change tested before it ships, all on AWS.
Next up: 🚢🚢🚢 All the innovation
Thanks @OpenSearchProj and @AWS for the support!
We tested 5 relevance tricks on 4 benchmarks.
Hybrid won on all 4. The reranker won on 3. BM25 tuning gained under 0.01 everywhere.
No single config wins. Every index needs its own decisions.
https://t.co/oCAE0LOnqE
Yet another reminder that we're living in the future.
This past week I built Haru, an AI assistant (inspired by Project Airi) to help keep my life organized. And she runs completely local. 🧵
Hot take for spooky season: "AI Search" for most companies means "we added embeddings"
Embeddings don't learn from clicks. They don't test ranking changes. They don't get better next week.
Same label. Different machine. 🎃
nDCG will likely remain the standard as far as humans are reading results and RCP-nDCG seems to fill a long time gap in the freshness of qrels.
Is it too late? We've found that for agent based search, much of our user base, using hits@n / recall@n has been sufficient.
I really like @cohere 's RCP-nDCG idea. It's always bugged me how many gaps are in human labeled datasets (even the old standards). Using an LLM to fill in the gaps is very natural: https://t.co/yF68bugWow
So we asked Jev if you should use Relevan for agentic search and the answer was 🔥
Being real though we tested out Jev to see if it could replace some parts of our relevance engine and were pleasantly surprised. It managed to be 100% accurate with our eval set.
Relevan is officially live! The beta is capped so you may get added to the wait list as you login. We are slowly letting people onto the platform.
If you're interested in becoming a design partner we still have a few slots open so reach out!
Join at https://t.co/oixuwaUPTy
Deadline coming up! Book your stay at the San Jose Marriott by August 30 to take advantage of the special rate for #OpenSearchCon North America attendees: https://t.co/8QijdKany0 See you in San Jose, Sept. 22-24!
Not registered yet? Save $100 now → https://t.co/GNSN6Akzhc
@jobergum This is one of the reason we're looking at Shots-to-Success as a new potential benchmark.
https://t.co/mgBzSNH4Kc
It helps measure if LLM's are stopping at the appropriate results in a multi turn search.
Relevan is officially live! The beta is capped so you may get added to the wait list as you login. We are slowly letting people onto the platform.
If you're interested in becoming a design partner we still have a few slots open so reach out!
Join at https://t.co/oixuwaUPTy
Each company has different data, users and goals. Most platforms give everyone the same tools.
Relevan is building an individual relevance loop for our customers. We measure quality, learn signals, test, and tune to your use case.
Beta tomorrow: https://t.co/oixuwaUPTy
Agent-first search isn’t a chatbot bolted onto an old console.
Relevan uses LLM-driven onboarding. Bring any agent, point it to our LLMs.txt, OpenAPI spec, and per-index skills, then let it configure, ingest, and search.
Beta opens Wednesday: https://t.co/oixuwaUPTy
Relevan wouldn’t exist without open source.
We’re built on @OpenSearchProj and @valkey_io—but open source shouldn’t be a one-way transaction.
We're excited to contribute code, knowledge, and improvements back.
Join the beta @ https://t.co/oixuwaUPTy
Agents break the old “query then click” feedback loop.
They branch, refine, and use results without clicking.
We’re working on that problem in the open through @o19s UBI standard: https://t.co/V2PGn1UfkZ