FeedLog

FeedLog is an open-source customer-feedback platform for collecting feature requests, voting and discussion, maintaining a public roadmap, and publishing changelogs. Unlike many feedback products where AI is mainly a summarization add-on, FeedLog uses AI in the workflow itself: semantic duplicate detection, widget-message triage and assisted changelog drafting.

Its duplicate system is technically documented. New feedback can be embedded into a 768-dimensional vector and compared semantically, so differently worded requests can still be surfaced as possible duplicates. FeedLog deliberately treats this as a ranking rather than an automatic merge decision: managers decide whether posts are actually duplicates, and vote counts are de-duplicated by voter when posts are merged. Without an AI key, the feature degrades to PostgreSQL trigram text matching rather than disabling the core feedback workflow.

The public roadmap exposes Planned, In Progress and Completed work, while the changelog supports drafts, published entries, update states, categories and reactions. AI drafting can use related feedback plus an operator-supplied description of what actually shipped; FeedLog's documentation explicitly warns that feedback is not itself a source of release facts and generated text should be reviewed before publication.

A product widget can be embedded through the @feedlog/widget package. The widget can convert a user's natural-language message into structured feedback or route the user toward support. FeedLog also supports JWT-based SSO handoff so existing product users do not need a second account flow.

FeedLog announced its open-source release in April 2026. The repository is MIT licensed and describes the current product as a Nuxt 4 / TypeScript application deployable on Cloudflare Workers, Vercel or Docker. The documentation says all three deployment paths run the same application.

Self-hosting requires PostgreSQL 17+ with the vector extension for semantic similarity. AI and several external services are optional. The AI layer is configured through an OpenAI-compatible endpoint, so teams can point FeedLog at OpenAI or at compatible gateways such as LiteLLM, vLLM or Ollama. This is useful for teams that want the application self-hosted while keeping model traffic inside their own network.

The architecture has meaningful operational caveats. If AI is not configured, semantic duplicate matching falls back to text matching; AI changelog generation reports that it is unavailable; and the widget's AI triage path cannot complete. Feedback created while embeddings are disabled needs a backfill before it can participate in semantic matching later.

FeedLog's current positioning is unusually simple: the open-source version is free and MIT licensed, and the hosted FeedLog Cloud changelog says the hosted service is also currently free to use with no pricing, usage limits or credit card. The hosted release states feature parity with the open-source product.

That makes FeedLog economically attractive today, particularly compared with feedback tools priced by admin seats or tracked users. But “free today” is not equivalent to a durable long-term commercial pricing model. Teams choosing the hosted service should therefore treat future pricing as an unresolved variable, while self-hosters should account for database, storage, mail and optional model/API costs.

FeedLog's first-party technical transparency is a strength. The help documentation goes beyond marketing and specifies semantic-search behavior, degraded modes, required database extensions, environment variables, SSO flows and widget APIs. The public repository adds code-level inspectability and reduces vendor-lock-in risk.

Independent evidence is much weaker. Searches across the reviewed source set do not currently establish a meaningful exact-product user-review corpus or mainstream independent test. The major software-review surfaces do not provide a usable FeedLog rating sample, and model benchmarks are not product-surface matches for a feedback-management application.

A few additional web references and user comments exist, but they are too small or weakly controlled to support a numeric quality signal. Absence of evidence is treated as unknown rather than zero; it lowers confidence in the conclusion.

Because FeedLog is open source, teams can independently inspect implementation and deploy a controlled pilot. That partially offsets the lack of third-party reviews for technical buyers, but it does not create evidence about long-term hosted-service reliability, support quality, ecosystem scale or adoption.

FeedLog is best suited to startups, indie products and small-to-medium SaaS teams that want a direct feedback-to-roadmap-to-changelog loop without enterprise-process overhead. It is particularly attractive when data ownership, self-hosting or cost predictability matters.

The AI capabilities are most useful when duplicate requests arrive with different wording or when a team regularly turns shipped work into release notes. The fallback behavior also means teams can use most of the product without an AI provider and add semantic features later.

It is less suitable today for enterprises that require a long independent track record, large-scale support evidence, extensive formal governance layers or a mature third-party integration/review ecosystem. Self-hosting also transfers operational responsibility to the adopter.