Gonka Labs - Monthly Report No.1 #1477
gonkalabs
started this conversation in
Show and Tell
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
Hey! Gonka Labs here.
On June 12th the community passed Proposal #74. That vote funded the next stretch of work - infra, ops, and shipping products for the ecosystem.
This is the first public progress report we promised in the proposal (monthly updates on GitHub Discussions). Short version of what landed so far:
1. OpenBroker
We already announced the launch here: OpenBroker - broker for brokers or Devshards as a service (Jun 23).
Quick recap: OpenBroker is managed DevShards-as-a-service. Register, deposit GNK, get an
obk-*key, point any OpenAI client athttps://api.openbroker.gonka.gg/v1(UI at https://openbroker.gonka.gg). No wallet whitelist dance, no running your own escrow rotation. GNK billing with no markup. Public stats at https://openbroker.gonka.gg/stats.Since launch (numbers as of Jul 18)
mainnet-v0.2.13-v3-post1, route prefix/devshard/v3, 50/50 sticky split behind the router. OpenBroker was already on the v1/v2 path at launch; we completed the network-mandated v3 cutover ahead of the v0.2.14 deadline so brokers do not fall off when classic/v1/devshardgoes awayOps + product work after launch
GET /v1/usageandGET /v1/usage/{id}so integrators can pull request-level cost/token data without scraping the UIFeedback thread (git discussions)
Thanks for the sharp measurements. Where we landed so far:
GET /v1/usage/{id}so you can see whether a request settled, what tokens/cost we recorded, and the FinalCost snapshot when the gateway returns it/v1/models/v1/models /v1/models - still on the polish list (model ids are also in docs / 400 text); not blocking chat completionsOpenBroker is also doing what the launch post promised: shake out new DevShard versions at production scale. v3 on our gateways was exactly that - migrate, smoke
/devshard/v3, keep MiniMax brokers online through the cutover.Links:
2. Pulse
Pulse is the live media + sentiment dashboard for Gonka (see it as a backbone and a part of proposal item named MTD / Marketing Transparency Dashboard - we shipped essential part that does data gathering first). It pulls public Gonka coverage from X, Instagram, YouTube, and the web into one place, scores tone via Gonka LLMs (
proxy.gonka.gg), and shows activity / reach / engagement next to GNK market data. Built so anyone can see what the ecosystem is saying without opening ten tabs.Live: https://pulse.gonka.gg
What shipped
Numbers (as of Jul 18)
Why this matters
Proposal #74 called out demand activation and making Gonka easier to follow for non-technical users. Pulse is a part of that "transparency layer" - a public pulse of coverage, tone, and creators, wired into gonka.gg and backed by Networks collective inference stack.
3. gonka.gg - bugs + product work
Explorer side of the proposal was "gonka.gg V2.0" - faster engine, better UI, mobile, accurate GPU/inference metrics. We are not done with that part of roadmap yet, but a lot of the painful day-to-day stuff is already fixed and several new surfaces shipped.
QA bug smash (~240)
We ran a dedicated desktop + mobile QA pass with an external unbiased audit, and closed on the order of ~240 tracker bugs (desktop tracker version into the 170s, mobile into the 60s - still an ongoing flow since we ship new features that are getting QA tested as well).
Rough shape of what that covered:
Shipped across a long streak of QA PRs and multiple levels of sanity checks. Not glamorous, but this is the difference between "works for some people" and "works for 70% of 10k+ people a day on phones".
Product / data updates on gonka.gg (since the vote)
Inference after DevShards (v0.2.12+)
devshard-pollerthat ingests off-chain session diffs into ClickHouse (the chain no longer emits per-inference txs the old way) and is properly aligned with V3 flow, and is optimized to not make heavy load for the network (overall produced load caused by data ingestion went down X27k)./network/inference(Inference tab) is driven from that pipeline - 24h stats, gateway traffic, timelines./devshard/v3and union live shard inventories so traffic does not silently drop when gateways switch).Network analytics
/network- weight distribution across AI models per epoch, with an all-epochs view (Redis-cached so it stays fast).Token holders
Infra behind the explorer (also see section 4)
Explorer: https://gonka.gg
Inference: https://gonka.gg/network/inference
Network: https://gonka.gg/network
Why it matters
Faster deploys, less risk of a bugfix taking down indexing, room to run the heavier pre-compute work from the V2 explorer roadmap.
4. proxy.gonka.gg - B2B inference + public status
proxy.gonka.gg is our OpenAI-compatible inference proxy on top of Gonka DevShards - API keys, balances, usage, docs, chat playground. Public API at https://api.proxy.gonka.gg/v1. Pulse (and other Gonka Labs apps) call the same stack for LLM work, so when we harden the proxy we harden half the product surface too.
Numbers (as of Jul 18)
Measured window below is since the Jul 4 dedicated-server migrate (usage DB moved with the stack - earlier Railway and even earlier Qupra DC history is not in this series). The product itself was already live before Proposal #74.
What shipped since the vote
api.proxy.gonka.gg) - moved off the flaky VPS / and too unreliable Railway path onto a tuned server with local Postgres, own DevShard gateway, Proxy edge for the API. Frontend stays on other host atproxy.gonka.ggwith rewrites into the API.mainnet-v0.2.13-v3-post1, route/devshard/v3) ahead of the network v0.2.14 deadline - blue/green cutover with a temp gateway + drain so in-flight MiniMax traffic survived the switch. Currently 32+32 (64 inflight escrows) active v3 escrows (MiniMax + Kimi)Why it matters
Proposal #74 called out proxy.gonka.gg as the B2B / analytics lane. This stretch was less "new landing page" and more "make the pipe carry real load": multi-billion token days, a public status surface customers can trust, and staying on the current DevShard protocol so keys do not die when classic
/v1/devshardgoes away. Same gateway lessons feed OpenBroker - one network path. More "b2b" work ahead.Links:
We are still heads-down on the rest of the Proposal #74 checklist - hardening infra under real load, polishing what already shipped, and building the next products on the list. More updates as they land. Stay tuned.
Links
Feedback welcome in the thread - what should we prioritize next month from the proposal list?
All reactions