OpenAkashic
A public library of methods and reported experience for people and agents.
Given a situation and goal, discover a method, inspect its conditions and sources, then attach what happened and discuss alternatives. The current direction is contextual know-how reuse, including open-ended work such as design, product experiments and career exploration.
The current portable implementation lives in public-methods/. It includes source-linked methods, public browsing, a clickable record graph, exact-version application reports, and an agent client. No API key is needed for public reads or local lexical browsing. Public demo writes remain disabled; self-hosted writes use a private operator credential.
Experimental: retrieval is candidate discovery, not expert endorsement or proof that the method solves the task. On 16 cross-domain development queries, hybrid retrieval found the intended first method 16/16, while on an earlier out-of-scope technical set it failed to abstain in 4/4 cases. Actual answer-quality improvement, human preference and same-quality billed-token savings are not established. Read the full results, not only a positive headline.
Closed Akashic personal memory and Open Time Machine are separate services. Private vaults, retired Samsung news data, credentials and operational databases are not part of this module. Older directories, installers, MCP claims and historical benchmarks elsewhere in this repository remain legacy material; they are not evidence of the current service's capabilities. Use the current module's instructions rather than old automatic installers.
Code license: Apache-2.0. Linked third-party originals retain their own rights; attribution and fixture scope.
Relationship atlas update
The public home page now offers global/local record navigation inspired by Graph Commons, Kumu and Obsidian. Follow method → exact source → another method, with separate application reports and opinions. See the atlas and the evidence report.
The new deterministic navigation workload passed 12/12 shared-source and 9/9 feedback lookups. A reusable flat URL index was smaller than the graph responses; general graph token savings are not demonstrated. The existing two-task design comparison observed no functional advantage. Record lookup success is not proof of better outputs. Reproduce the workload and inspect the preserved unfavorable results in public-methods/benchmarks/.
