Reported issues for logr

Pod holds 9 of 9 problems reported by people outside the maintainer team. Issues filed by the project's own owners, members and collaborators are excluded entirely — a maintainer's release checklist is not a warning to a prospective user.

Back to logr.

Most discussed

Expose each profile as an MCP endpoint (/[username]/mcp)

Summary

Expose every logr profile as an MCP (Model Context Protocol) server. Markdown context remains the universal read-only fallback; MCP adds selective resource discovery, querying, and eventually authenticated owner mutations.

Relationship to Markdown context

The Markdown context work is tracked separately. MCP should reuse the same normalized ProfileDTO and complete Markdown generator rather than introduce a competing source of truth.

Suggested public resource:

`logr://pr

Read the thread · 2026-06-13 · open · 2 comments

pgvector RAG grounding for visitor chat

Summary

Currently, the visitor chat (#9) grounds its responses by injecting the context into the system prompt. As profiles grow (hundreds of events, long lore), this hits token limits and dilutes retrieval accuracy for specific questions.

Introduce pgvector to store embeddings of events/lore, enabling semantic retrieval (RAG) for the chat. This allows the AI to fetch only the most relevant events for a specific question ("What did they build in 2021?") rather than scanning the whole his

Read the thread · 2026-06-24 · closed · 1 comment

Cross-profile verified collaborations

Summary

The current trust model (#5) focuses on verifying events from external sources (GitHub, domain, wallet). This issue adds social verification by allowing users to tag other logr profiles on their events. If both parties confirm the collaboration, it creates a verified edge in a logr-native social graph.

This is the strongest compounding mechanic for logr: every tag is an invite, and every confirmed edge makes the graph richer and impossible for a static clone to replicate.

Pro

Read the thread · 2026-06-24 · closed · 1 comment

Verification & vouch model for credentials (claimed vs verified)

Summary

Introduce a claimed vs. verified distinction for profile data, plus a lightweight vouch mechanism. Once profiles are agent-authored and agent-queried (see #4), trust becomes the whole game — an agent should be able to tell what a user merely typed from what is actually confirmed.

This is a feature LinkedIn structurally cannot ship well, and it's what makes an agent trust a logr profile over a scraped webpage.

Marked for later — sequence after MCP (#4) and onboarding

Read the thread · 2026-06-13 · open · 1 comment

Distribution: Claude Code plugin — bundle the logr MCP server + usage skill

Summary

Package logr as a Claude Code plugin so logr becomes a one-command install for Claude Code users: the plugin wires the global MCP server (#58, shipped in #59) automatically and ships a small skill teaching agents how to use it well.

A skill alone can't establish an MCP connection, and a bare URL still asks the user to run claude mcp add … — a plugin bundles both halves. This is distribution channel 3 from the #58 follow-up (share modal ✅ shipped in #59; official MCP registry

Read the thread · 2026-08-07 · open · 0 comments

Global MCP endpoint (/mcp): query any profile + OAuth 2.1 owner writes

Summary

Serve one global MCP server at /mcp (Streamable HTTP, stateless) so any configured agent can query any public logr profile — and, once authorized via OAuth, maintain the owner's own profile.

This supersedes the per-profile /[username]/mcp design in #4: MCP clients connect to servers via static configuration, so an agent that encounters a new person mid-conversation can't attach a new per-handle endpoint on the fly. "Query any profile" therefore requires a single global e

Read the thread · 2026-08-06 · closed · 0 comments

pgvector RAG grounding for visitor chat

Summary

Replace the current full-profile context injection in visitor chat with pgvector-backed semantic retrieval over the profile's events. Grounded chat currently stuffs the entire llm.txt into the system prompt — this works at small profile sizes but has three failure modes:

Token ceiling — at a few hundred events the profile exceeds the system prompt budget, and the chat silently truncates or degrades quality.
Signal dilution — questions like "what did they ship in

Read the thread · 2026-06-24 · open · 0 comments

Epic: logr as the identity layer for the agent era

Thesis: A resume is a document you write. logr is context that gets read — by a recruiter's eyes for 6 seconds, or by an agent for 6 milliseconds. The defensible wedge isn't "better portfolio tooling," it's becoming the read / query / verify layer for human identity in an agent world.

We already let people log once and produce three things: a human timeline (/[username]), a machine-readable llm.txt (/[username]/llm.txt), an

Read the thread · 2026-06-13 · open · 0 comments

Most recent

Socials field duplicates entries and breaks links when format isn't exact

Summary

The socials field on the profile duplicates its entries on every save and renders broken, icon-less links whenever the input isn't formatted exactly as Label https://url.

🐛 Originally reported by @Ishwarpatra in #1 — thank you for catching this and digging into the root cause.

Steps to reproduce

Open the dashboard → profile → socials.
Enter a link that isn't exactly Label https://url — e.g. paste a bare URL (github.com/name), or copy the field hint literal

Read the thread · 2026-06-11 · closed · 0 comments

The remaining reports are on the project's issue tracker.