01 / Process
How these notes are made
After code lands, we reconstruct the story from its commit, diff, tests, and runtime evidence. The result becomes a durable technical document for users and contributors.
OpenMake Engineering Log
Not a changelog echo. Each entry explains the problem, the design decisions, the implementation boundary, the evidence, and the trade-offs behind a shipped OpenMake change.
01 / Process
After code lands, we reconstruct the story from its commit, diff, tests, and runtime evidence. The result becomes a durable technical document for users and contributors.
02 / Team
OpenMake Team is the community identity; openmake_llm is the software project; OpenMake is the product name. Non-developer maker riskpw leads openmake_llm through vibe coding, and professional developer rocky supports its development.
Engineering deep dive

How OpenMake changed large agent-task attachments from one oversized request into an authenticated, resumable-by-chunk protocol while preserving the existing task file contract.
Read the engineering note2026 / W04–W32

The current partial week strengthened orchestration reliability and frontend-backend parity, then completed a path that sends attachments beyond Cloudflare's per-request limit in chunks and analyzes scanned PDFs with no text layer. Small documents use a multi-page native OCR fallback, while large documents move into the sandbox for targeted page OCR.
Read the engineering note
The final July week shifted from feature accumulation toward architecture. Provider contracts, desktop boundaries, MCP behavior, operations, and a LiteLLM and vLLM topology were reviewed to keep orchestration ownership inside OpenMake.
Read the engineering note
Late July connected autonomous work to real project and identity boundaries. Agent-task Git operations, OAuth paths, source-wide audits, Cloudflare migration work, release documentation, and desktop recovery were handled in parallel.
Read the engineering note
Model assignment became explicit while the product gained new gateways and stronger chat behavior. Discord integration, provider roles, execution planning, routing measurements, and self-learning evaluation were advanced with visible trade-offs.
Read the engineering note
The second July week combined implementation with deep inspection. MCP catalog additions, agent-task design, self-learning measurements, frontend audits, and interactive log tooling exposed both progress and unresolved failures.
Read the engineering note
The transition into July consolidated web execution, artifacts, and agent tasks, then returned to live chat and authentication behavior. Exact-cwd Claude sessions resume in the archive during this period.
Read the engineering note
The busiest June week pushed execution beyond simple conversation. Task sandboxes, browser and web paths, artifact handling, attachments, MCP tooling, and chat orchestration became a connected runtime.
Read the engineering note
As capabilities grew, this week strengthened their guardrails. Security fixes, API-key controls, deep-research behavior, caching, design consistency, and usage visibility were improved together.
Read the engineering note
Research and answer quality became the focus. Deep Research, evaluation hooks, prompt work, chat behavior, artifacts, and agent-task refinements were developed as one evidence-oriented answer pipeline.
Read the engineering note
The first June week established agent tasks as more than a chat mode. Task execution, administration, routing, and a broad interface pass created a dedicated surface for longer autonomous work.
Read the engineering note
Late May centered on artifacts, chat routing, sockets, and structural cleanup. Generated outputs gained a clearer lifecycle while routing work reduced the gap between model decisions and the interface.
Read the engineering note
This was the largest weekly commit cluster in the audited period. MCP ingestion, skill creation, agent ingestion, model catalogs, navigation, and GDPR work turned isolated capabilities into an administrable platform.
Read the engineering note
After the large provider and UI sprint, the next week was quieter and review-heavy. Security, environment handling, and refactoring reduced the risk introduced by the previous volume of change.
Read the engineering note
Development accelerated sharply in early May. Provider connections, model selection, brand presentation, and the main chat interface were rebuilt together so the growing backend remained understandable to users.
Read the engineering note
The final active April week closed storage fixes and small feature gaps. No matching Claude transcript was recovered, so this note stays anchored to Git rather than reconstructing an AI-assisted narrative that cannot be proven.
Read the engineering note
Security controls became concrete implementation work. Content Security Policy, WebSocket behavior, CSRF protection, storage rules, and response headers were tightened across the public and authenticated surfaces.
Read the engineering note
April work continued in Git even though no matching local Claude Code transcript was recovered. The repository shows infrastructure, configuration, and operational documentation changes during this week.
Read the engineering note
The turn from March into April was deliberately smaller. Documentation, configuration, and maintenance work consolidated the previous hardening sprint before the next infrastructure phase.
Read the engineering note
The codebase shed older abstractions while new agent and routing behavior matured. Removing legacy A2A paths reduced ambiguity and made the remaining execution model easier to reason about.
Read the engineering noteOne of the busiest March weeks produced a wide set of fixes and refactors. API-key flows, internal memory, agents, RAG, MCP behavior, and user-facing defects were tightened as one reliability pass.
Read the engineering noteA smaller week concentrated on fixes, provider behavior, and integrating parallel branch work. The goal was not a new headline feature but keeping the fast-moving platform coherent.
Read the engineering noteEarly March focused on making richer answers operable. Firecrawl integration, cache behavior, MCP and AI feature toggles, ES module cleanup, Mermaid, and KaTeX were reviewed and stabilized together.
Read the engineering noteOpenMake started growing beyond basic chat. Skills, memory, schemas, upload security, MCP controls, and deeper testing established the components needed for an extensible AI workspace.
Read the engineering noteAttention shifted to the boundaries users hit first: sign-in, anonymous sessions, route behavior, and server reliability. Refactoring and bug fixing turned early features into a platform that could survive ordinary use.
Read the engineering noteThe second February week expanded the initial platform, refreshed project documentation, and prepared the codebase for faster iteration. The work was still foundational: make the system understandable before adding more surface area.
Read the engineering noteThe first repository commit landed on February 3 as OpenMake LLM 1.5.0. This is the point where the January design exploration became an auditable codebase milestone.
Read the engineering noteThe last week of January moved from system mapping into a concrete router-conflict implementation session. Provider flow and responsibility boundaries became explicit, but the public repository history had not started yet.
Read the engineering noteThe January archive begins with reconnaissance rather than a repository milestone. The sessions mapped the existing OpenMake surfaces, deployment assumptions, and routing conflicts that had to be understood before implementation could start.
Read the engineering note