
Artificial Superintelligence Alliance · 2026-11-13 15:31 UTC
ASI Product & Development Update | Edition 19 This nineteenth edition of the biweekly ASI Alliance Product and Development Update covers progress from October 29 to November 11, 2026. During this period: ASI:Chain development included the release of Wallet SDK 2.1.0 and the redesigned Web Wallet 0.2.0. Support for exact token amounts expanded across several services, and load testing identified ways to limit how much data nodes keep on disk. The Omega team published v0.1.24 as a pre-release. Other development improved how plugins declare dependencies, how agents manage longer activity histories, and how reasoning settings reach supported model servers. ASI:Chain Node Infrastructure As a blockchain processes more transactions, its records naturally take up more disk space. Load testing found that nodes also keep older versions of balances and other chain data, and that some data used to process transfers stays on disk after transfers complete. The team is evaluating how much history regular nodes need to keep while preserving the full record on archive nodes. It is also looking at how transfers can leave less data behind. In a separate test on a read-only node, limiting how many past reports the node kept in its cache prevented that cache from growing. A restart issue that could keep a node disconnected when its initial peer was offline was fixed. Changes under review add checks to snapshot recovery so a node can reject data from a different chain and stop retrying when the mismatch is permanent. New tests examine how nodes behave when validators have different voting weights or when the peer they first connect to goes offline. Amount Precision ASI:Chain records token amounts as whole numbers of the token's smallest unit. During earlier staking tests, a reward calculation exceeded the limit of the old number format. The wider 256-bit format supports much larger balances, transfers, and rewards while preserving all 18 decimal places. Following the bridge tests reported in Edition 18, support for the wider format expanded across the chain's services. The node's HTTP API preserves large amounts without rounding, the Rust client can sign transactions that use the wider format, and the faucet supports configurable token precision. Indexer testing verified that it could repair historical transfer records and recover after a restart. Work on the Wallet SDK and Web Wallet is adapting balances, transfers, and fees to the selected network's token precision. This lets the same wallet interpret amounts correctly on networks that use eight or eighteen decimal places. The SDK's revised conversion rejects amounts that cannot be represented exactly in the token's smallest unit. The wallet compatibility changes are still in development. Wallet SDK and Web Wallet Wallet SDK 2.1.0 was released on npm on October 28, after the reporting cutoff for Edition 18. See the release notes. The update reserves funds before a transaction is submitted, releases them if the node rejects it, and keeps them reserved if the outcome is uncertain. It also validates account entries before importing a keyfile and distinguishes wallets created from the same recovery phrase with different account-generation settings. Setting the auto-lock delay to zero requires the password for every signature. Web Wallet 0.2.0 is the first release built on the SDK. It supports multiple accounts from one recovery phrase, encrypted keyfile import and export, signing sessions with auto-lock, and a redesigned transaction history. Wallets from the previous version must be restored using a recovery phrase or imported by private key. The Bridge page is present with its lock action disabled, and transaction history is limited to 50 records. Further development added a balance check immediately before submission so the wallet uses a fresh balance when deciding whether to send a transaction. In the SDK, an encrypted backup of all stored wallet data is under review. The import validates the entire backup before writing data, so a damaged file or incorrect password fails before existing storage is touched. Network Operations and Cost Accounting Updated node, faucet, and command-line wallet images were published for testing. The test environment was adapted to cover more scenarios under the revised cost-accounting rules. Compatibility changes for the node and the services that submit, index, or display transactions are still being integrated. Omega Release and Memory Reliability The Omega v0.1.24 pre-release fixes how the agent handles memory errors. When users switched to a model incompatible with their existing memory store, Omega could report that it had saved or found a memory even though the operation had failed. The fix reports the failure and identifies the settings involved. To keep using an existing memory store, users should keep the provider and model that created it. Release tests verified that upgrading Omega itself from v0.1.23 preserved the existing memory store through subsequent restarts. Exporting and reimporting the store preserved the same data. The team tested the one-line installer, installation without Docker, and startup without administrator privileges on both Arm64 and x86-64 builds. Installation and Model Configuration The installation guide was updated with current launcher commands. Memory migration warnings link to a fixed version of the relevant instructions, and users can configure Omega to suppress the startup version message. An opt-in setting lets users tell supported OpenAI-compatible model servers how much effort to spend reasoning before replying. Previously, requests could omit that setting and use their entire token budget without producing a reply. Automated tests also check that requests format tool definitions correctly, that reloading a workflow does not duplicate tools, and that an idle agent does not send repeated requests. Plugins and Agent Context Plugins can list the Python packages they need to run. If plugins require incompatible versions of the same package, developers can catch the conflict while building the agent. The plugin documentation explains how to package tools the model can call and how to define skills with a SKILL.md file. A configurable limit was added to control how much task history Omega keeps in its working context. When the history reaches that limit, Omega summarizes earlier activity, keeping the original instructions, key decisions, and outcomes in view. The agent can still look up the full details in the archive. Omega’s context classifier was updated to use the language model configured for the agent. IRC Messaging Omega’s IRC integration was updated to keep emoji and Cyrillic characters intact when splitting long messages to fit the server’s size limit. If the connection drops mid-message, Omega resumes sending the unsent parts after reconnecting. Security and Reasoning Evaluation As part of its ongoing security evaluation, the Omega team selected AgentDojo to test how agents respond to misleading instructions embedded in external content. The benchmark measures whether agents can still complete their assigned tasks and how often those instructions change their behavior. The team is also exploring how to test whether agents base their reasoning on what they observe in their environment. Visit the ASI Alliance website to learn more. See you in two weeks. The ASI Alliance Product & Development Team
Loading the story on Tree News…