Find your next read across module mechanisms, architecture and engineering practice.
Links marked “Chinese original” lead to articles without a published English edition. Published translations retain the original article identity.
Articles must match all selected tags. Changing tags or order returns to page 1.
Created means this language edition’s first recorded version on the site’s main branch, following Git-recognized renames—not the original writing date. Updated follows visible content changes. Dates: UTC+8; unknown dates sort last.
Showing 1–20 of 123 articles · Page 1 of 7
Go to results ↓A field-tested architecture case study of Core API, Product API, and Background Worker: measured lifecycle isolation, foreground capacity handoff, crash recovery, and local transport identity, without claiming production SLA gains.
What changes when ordinary research becomes a long-running task? Follow a three-fund comparison through concrete integration steps for request assembly, budgets and finalization, material rereading, checkpoints, and report revisions, with links to the existing implementation guides.
Start with the background and purpose of specifications, then follow file ingestion through principles, design, templates, behavior, verification, archiving and template maintenance—without copying project-specific ceremony.
Follow a fund-comparison conversation down to tasks, runs, model turns, request attempts, and tool actions. See which identities continue through replies, process recovery, and scheduled execution—and use those distinctions to design and debug an agent.
Expert profiles define a method, subagents execute bounded work, reviews inspect a report, and judges evaluate results. Follow one fund report to distinguish their inputs, outputs, and authority before adding another model call.
Continue the first workshop by connecting selected Web sources, specialist methods, and report review. Freeze the evidence, check citations, and decide what to change without overwriting the original report.
An Agent run may contain multiple model requests, failover, child work, and recovery. This article explains actual execution identity, effective behavior inputs, three-way comparability, and the explicitly limited reuse of historical local evaluation results.
Follow one document comparison to distinguish instructions, current input, retrieval and lasting memory. Use Open ETF Agent's offline examples to decide which information needs attention when an answer goes wrong.
Follow one product-guide read through tools, function calling, MCP and Skills. Distinguish executable capabilities, call representation, connection protocols and working methods using public code.
Return to a document comparison after closing the program. Distinguish conversation history, current input, lasting memory and execution checkpoints to understand what viewing the past and resuming work each require.
A size limit has been exceeded, a 401 has arrived, or a connection has timed out. Why is the failure still not reaching the caller? Examine asynchronous waits, error classification, and shared slots to decide when cleanup belongs in result settlement and what evidence is still missing between requesting cancellation and reclaiming resources.
Compare OpenAI, Anthropic, Google, and open-source research practices, then examine ETF Agent’s resource, parallel-evidence, and delivery choices through its implementation and six matched experiments.
How do evaluation sets connect requirements, development, real-task acceptance, and operational feedback? External practices and ETF Agent's current 83 questions explain case selection, engineering closure, and research-quality evidence.
Why ETF Agent adopted and then retired LangGraph, how 26 LangGraph concepts map to concrete research-runtime responsibilities, and where persistence, human interaction, and publication contracts diverge.
From behavior contracts, understandable changes, and checks that can detect mistakes to review, delivery, and continued evolution: how to make sound quality judgments while keeping verification and maintenance affordable, with examples from our open-source work and established industry approaches.
Trace one fund-research task through Chat, Copilot, Agents, and long-term assistants: context construction, execution control, tool calls, durable state, and recovery, grounded in Open ETF Agent v0.24.
How can a fund research agent preserve the provenance of evidence as it moves across websites, PDFs, calculations, subagents, and historical reports? Follow one research task to distinguish traceability, method applicability, and sound investment judgment.
A bounded review of real-model fund reports shows how cash reconciliation, incomplete holdings coverage and ETF trading rules can invalidate claims even when the report looks complete; learn to assess what each claim is entitled to say.
Learn who saves checkpoints, when they are triggered and how one Run updates them. Follow the same task through LangGraph, Microsoft Agent Framework, Temporal and ETF Agent to compare continuation models and ownership.
From optional-note vetoes to termination ownership, Spec, and History: why safeguards can amplify local failures and how engineering can reduce repeated control and evolution debt.
Articles reflect their own evidence and context; design intent, bounded verification and actual delivery are different claims.