(outdated) Proposal #74: Gonka Labs - Monthly Report No. 2 #1617
Closed
nsvdev
started this conversation in
Governance Proposal Reports
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. Last update was Monthly Report No.1 on Jul 18.
Report #1 was the launch month: OpenBroker and Pulse went live, the explorer got a big QA + inference pass, and proxy.gonka.gg moved onto dedicated infra under real load. This month is what we added on top. New transparency dashboard, a real agent workspace on the proxy, a third model on our own DevShards, and the ops that show up when hosts leave mid-epoch or a disk fills up.
Short version of what landed since Jul 18:
1. proxy.gonka.gg. Workspaces + third model + agent compat
proxy.gonka.gg is our OpenAI-compatible inference proxy on top of Gonka DevShards. API keys, balances, usage, docs, chat. Public API at https://api.proxy.gonka.gg/v1. Report #1 was the dedicated-box migrate, public
/status, and DevShards v3. This month we put a real product on that pipe, and we made agents plus a third model work at the same scale.Short version
$ref/$defs, emptytool_call.id, MiniMax content shape, leftover client knobs/statusnow shows mid-epoch confirmation PoC, so a capacity dip is not mistaken for an outageNumbers (since Jul 18)
What shipped
Workspaces are not just a playground chat. The agent writes docs, slides, sheets, and code in one session on Gonka inference. Pick a surface at https://proxy.gonka.gg/workspaces and share the finished space.
In
/chatyou can connect any MCP (search, external services, your own connectors). The agent calls them in-dialog. Attach images, PDF, text, HTML, XML, CSV, XLSX, etc.The Proxy parses files sent by user (images, pdf, xlsxs, etc) and sends the model the meaning (if user enables "File Ingestion" option in the Account settings). Attachments in Workspaces stay free.
On our own DevShards: DeepSeek as the third model, same 32-escrow target as MiniMax and Kimi. Escrow refill-to-target (settlement stays off). Empty-escrow 502s are no longer the failure mode.
/statusshows network confirmation PoC when hosts leave inference mid-epoch. That is a rate limit, not a Proxy outage.Compat, because Cline / agents suddenly 400'd: Gonka hosts tightened tool-schema validation. We kept the old repairs and added request rewrites so clients still work. Inline
$ref/$defson tools andresponse_format.json_schema. Fill emptytool_calls[].id(and matching tool results). MiniMax content becomes[{name,type,text}]on every role.reasoning_effort: maxbecomeshigh. We dropextraBody/json_mode/output_config/fallbackbefore upstream.Why it matters
Proposal #74 listed proxy.gonka.gg as the B2B / analytics lane. Report #1 made the pipe carry real load. This month is the product on that pipe: a shareable workspace, agents that do not 400 on strict hosts, and a third model with enough escrow that a mid-epoch host exit looks like a rate limit, not an outage. Same gateway lessons still feed OpenBroker.
Links: Product · API · Status · Workspaces
2. PTD. Proposal Transparency Dashboard
Report #1 shipped Pulse as the first slice of the proposal's transparency item (then called MTD). PTD itself was built this window and launched in August.
Quick recap: every passed on-chain proposal is listed automatically. A team can claim a page (contact required, moderator review) and then post tasks, rich updates, and discussion. Core-team GitHub category reports are ingested automatically and show for claimed and unclaimed proposals. No login to read.
Live: https://ptd.gonka.gg
What shipped
Product built late July (email + GG Wallet auth, proposal cache from rpc.gonka.gg, claim / tasks / append-only updates / discussion). https://ptd.gonka.gg went live with TLS on Jul 31.
Early August we reworked on-chain tasks on the claim form, put the full proposal text on the initiative page (claimant can edit it), and sent claims through a moderation queue. Images work in the rich editor.
Aug 6: ingest from Core's Governance Proposal Reports (
Proposal #<n>: …threads). Public/reports/[id]. No auth, no claim wall. Home shows the latest 5 reports (preview image + excerpt) plus an all-reports grid.Aug 13: OG title / description / image so Telegram (and others) get a proper card when someone shares ptd.gonka.gg or a report link.
Why it matters
Proposal #74 asked for a transparency layer so people can see what funded work is actually doing. Pulse was the media slice. PTD is the governance one: initiative pages and the reports we (and others) post on GitHub, readable without a login.
Links: PTD · Reports · GitHub source
3. gonka.gg. Explorer + inference + holders
Report #1 already had the inference tab, Model Weight Share, GPU / reward charts, and the holders page, plus the ~240 QA smash. This month is what we added or fixed on top of that.
Short version
What shipped
Public DevShard aggregates on the API so partners do not scrape the UI. New broker added in the gateway panel. Live Devshard Flow "participants" was capped at 12. Uncapped it. Readable tx type labels, copy toasts, epoch countdown tooltip in the header. MCAP / FDV on the ticker. Supply + accounts rows put back in the price tooltip. AI assistant reliability / live-data pass (assistant was already there).
Chinese localization finished across the explorer. Wallet page: account details + transaction summary.
Holders: account bubble map, new/active activity charts, holders chart. Account totals made cumulative (dust transfers still count in the total).
Honesty fixes elsewhere: escrow inference duplicates (diff nonce vs id), stale ghost ML nodes no longer inflate GPU counts, reward calculator showing 0% GPU weight coverage, proposal detail voter count (QA: proposal 95, "3 vs 12"). Inference tab: Confirmed/All on 24h gateway traffic; 24h model tokens table. Model Weight Share recounts hosted PoC, not confirmation weight (the chart was already in report #1).
Phone: homepage blocks, header, blocks/rewards widget. GPU page on phone defaults to top-4 + a latest-epoch mix instead of 12 overlapping lines. Inference tab no longer overflows the viewport on a phone.
Why it matters
Explorer side of the proposal was gonka.gg V2.0: faster engine, better UI, mobile, accurate GPU/inference metrics. Report #1 closed the painful day-to-day bugs. This is the next layer of that.
Links: Explorer · Inference · GPUs · Holders · Inference map
4. OpenBroker
We already announced the launch in discussion #1363 and covered it in Report #1: dedicated box, dual-gateway router, DevShards v3,
obk-*keys, GNK ledger billing,GET /v1/usage/{id}, public stats. This month is capacity + ops + broker UX under real relay load.Quick recap: OpenBroker is managed DevShards-as-a-service. Register, deposit GNK, get an
obk-*key, point any OpenAI client at https://api.openbroker.gonka.gg/v1 (UI at https://openbroker.gonka.gg). GNK billing with no markup. Public stats at https://openbroker.gonka.gg/stats.Numbers (since Jul 18)
What shipped
Third model: DeepSeek-V4-Flash-0731 minted on both production gateways (live Aug 14). Docs / status / canonicalize updated.
Escrow maintainer now refills MiniMax / Kimi / DeepSeek to per-model targets on v2 + v2b independently. We tightened hard opens/run + opens/day caps and added a min operator-wallet floor.
DeepSeek clients get
reasoning+reasoning_content(alias so contract clients see thinking fields).Deposit dashboard now says the address is custodial, not personal wallet, with explicit steps. Password-reset feature added.
Ongoing: keep MiniMax / Kimi / DeepSeek escrow pools topped for dual-GW HA. proxy.gonka.gg Secondary / joingonka fallback accounts keep using OpenBroker as their overflow path. Common sense.
Why it matters
OpenBroker is still doing what the launch post promised: shake out new DevShard versions and models at production scale. DeepSeek on both gateways, refill-to-target, honest billing when final-cost is missing, plus deposit / reset UX so brokers can actually get back in when something breaks.
Links: Product · API · Stats · Launch post
5. Pulse
Report #1 launched Pulse with X / Instagram / YouTube / web, 24h charts, sentiment on Gonka LLMs, and the public read API. This month is more sources, cleaner signal, and a couple of community surfaces.
Live: https://pulse.gonka.gg
Numbers (as of Aug 18)
What shipped
Reddit + TikTok collectors. Noise / Top Creators filters so homonyms do not pollute the feed. Off-topic X "gonka" homonym filter and UTF-16 emoji sanitization in ingest. TikTok Top Creators noise filter so junk accounts do not dominate the board.
Telegram community pulse bot + discourse "talking about" digest (EN + RU), with sentiment chips / topic bars on the community panel.
More web / press: curated sources, Medium RSS, original-article discovery. Mobile + feed QA, creator profile links, GNK price archive fallbacks.
Why it matters
Proposal #74 asked for demand activation and making Gonka easier to follow. Pulse is still that layer. Two more public sources, a Telegram surface if you do not live on the dashboard, and reach numbers that do not jump when Instagram resets a counter.
Links: Pulse · API: https://api-production-960a.up.railway.app/api (stats / feed / news / creators / telegram / discourse)
6. rpc.gonka.gg
Public RPC / LCD gateway. Report #1 mentioned it in the footer. The free / no-key pool, Hyperfusion + 6block, JSON-RPC failover, CosmWasm limits, and IP scrubbing shipped in June, before that report. This month is gRPC, a live-host cutover, and not hanging wallets when a Feather falls off tip.
Short version
rpc.gonka.gg:9090(TLS)/healthno longer hangs if one sub-service is deadWhat shipped
Mobile: full mobile layout shipped.
Some RPC nodes in the pool lagged in late July. We drained the lagging nodes from the community pool, rerouted traffic, then brought them back in when they caught up.
Cosmos gRPC is forwarded on
rpc.gonka.gg:9090(TLS at the edge, our Feather nodes behind it). Weight / drain when a Feather is in recovery after an upgrade.Live gateway cut over to a new box. Moved to better servers. ClickHouse + tx scanner follow the archive host. One Feather decommissioned. Pool still serves as it was, just faster and more stable. Health probes are time-bounded so a dead scanner cannot freeze
/healthor the homepage "Powered by" strip.Why it matters
Wallets and the explorer only look healthy if the public RPC stays on tip. This stretch was: move the box, add gRPC, do not take Keplr down while we do it. So we did.
Links: Gateway · Docs · Health · gRPC:
rpc.gonka.gg:9090· Access (optional)7. G-Meter
This month is ops hardening + QA + DeepSeek / OpenBroker / AI status.
Live: https://meter.gonka.gg
What shipped
/comparelinked in the top nav. Explained aggregate "failed tests %" vs latest-run broker list (history includes older underpayment failures).Front QA: Limits / Providers / pricing UI; live
/v1/modelsprices, shared price-scale domain, CIS language on compare. Mobile probe-log + providers layout.DeepSeek-V4-Flash-0731 added to the active probe lineup (catalog / probes / brokers). Trends filter dedupe. Dropped DeepSeek probes on brokers that do not list it (Hyperfusion / Gonkarouter).
OpenBroker (
api.openbroker.gonka.gg) added as OpenBroker by Gonka Labs. DeepSeek viaproxy.gonka.ggwrites cached network + broker AI summaries (refresh after probes). At a glance opens the broker note from a message-icon modal, with plain-language notes like "billing unavailable".Why it matters
G-Meter is how anyone can see whether a broker is actually serving, at what price, without trusting a landing page. Adding DeepSeek and our own OpenBroker to that lineup, and not dying during a host cutover, is the point of the page.
Links: Dashboard · Compare · Get started
Still working to make Gonka Labs products better while creating new onsed. Hardening infra under real load, polishing what already shipped, building the next products on the list. More updates as they land.
Links
Feedback welcome in the thread,
Gonka Labs team.
All reactions