## User Feedback Analysis — 2026-02-24 (based on Feb 21–23 Discord + Feb GitHub activity) ### 1) Pain Point Categorization (top recurring 5–7) #### A. Documentation — v2.0.0 onboarding + “where is the truth?” gaps (high frequency, high severity) **What users report** - Confusion about which branch/version to use: *v2.0.0 is recommended* but “has less documentation” (Discord, Feb 21). - Missing/hidden examples for key capabilities: HTTP server wrapper exists “in Milady with numerous examples,” but not clearly discoverable (Discord, Feb 22). - Database setup and migration unclear: - Users ask how to use **pglite vs Neon Postgres** and whether data migrates (answer: no auto-migration) (Discord, Feb 21). - Removal of v1.4.x → v1.6.x auto-migration raises upgrade anxiety (Discord, Feb 22, PR #6521 context). **Rough quantification (this sample):** ~4 of ~11 distinct Discord Qs/threads (~36%) are “how do I set up / what version / where are examples” style onboarding/documentation blockers. --- #### B. Technical Functionality — token/context bloat and duplicated calls (high severity, medium frequency) **What users report** - Context/token blow-ups: - Regularly hitting “**200k token limits**” with many plugins; bootstrap providers + evaluations called out as major contributors (Discord, Feb 22). - MCP loading + METATOOL search actions discussed; root cause suspected to be **recent messages/reflections collecting too much data**, not actions (Discord, Feb 23). - GitHub bug: **URLs trigger duplicate LLM calls** (processed as both text and attachment), doubling cost and producing duplicated output (Issue #6486, OPEN). **Rough quantification:** ~2 of ~11 Discord items (~18%) plus at least one high-impact open GitHub bug point to cost/call amplification as a core reliability/cost problem. --- #### C. Integration — provider/plugin fragmentation and compatibility gaps (medium frequency, high leverage) **What users report** - Plugin confusion for image generation: - Users ask if the older `plugin-image-generation` is still OK; told to use `plugin-bootstrap` image action instead (Discord, Feb 23). - OpenAI-compatible endpoints not supported: - Feature request: configurable **custom OpenAI endpoint URL** to use OpenAI-compatible services (e.g., SiliconFlow) (Issue #6490, OPEN). - Local model integration instability: - Embedding failures on Linux in `plugin-ollama` (Issue #17 in plugin-ollama; referenced in weekly/monthly summaries). --- #### D. Community / Trust & Safety — migration confusion + scam risk (high frequency, high severity) **What users report** - Token migration confusion, missed deadlines, and value loss; distrust when asked to provide wallet addresses (Discord Feb 23 raw summary). - Scam warnings are repeatedly necessary: “avoid DM links” and fake ticket systems (Discord, Feb 23). - Confusion about which token receives an “upgrade” (ElizaOS vs $milady.ai) remained unresolved (Discord, Feb 22). **Rough quantification:** ~4 of ~11 Discord items (~36%) relate to token identity/migration/trust questions (migration process, token availability, “which token is upgraded,” scam avoidance). --- #### E. UX/UI — unclear “how to do X” pathways (medium severity, lower frequency but persistent) **What users report** - Users can’t find where to open a ticket / correct support path (“missed migration” → can’t find channel) (Discord, Feb 23). - Webapp streaming UX issue in #6486 (duplicate text in SSE stream) blends UX + backend behavior. --- #### F. Performance — baseline overhead from bootstrap providers/evals (high severity, medium frequency) **What users report** - “Baseline” token consumption attributed to bootstrap providers + evaluations; desire to reduce default prompt footprint even before user adds plugins (Discord, Feb 22). --- ### 2) Usage Pattern Analysis (actual vs intended) **Observed usage patterns** - **Agents as crypto/trading automation and commercial endpoints**: community and core-dev focus heavily on Babylon/Hyperscape/OTC agent, exchange listing BD plugin (SolCex), wallet/trader concepts (Spartan/DegenAI). This suggests many users treat elizaOS as an *agent commerce framework* more than a generic assistant toolkit. - **Plugin-first building**: users quickly assemble many plugins, then hit context/token ceilings (200k token limit report). This indicates real-world deployments are “capability stacks,” not minimal agents. - **v2.0.0 adopted despite doc debt**: users are being told v2 is “the future direction,” so they accept instability/documentation gaps and ask for examples. **Emerging/unexpected use cases** - **Non-technical end-user agents** (Reservation/Travel/DB analysis) showcased by DarmaStef (Discord, Feb 22). This is a strong signal for templates and opinionated starter kits (not just low-level framework docs). - **Cross-language runtime interest** (Rust/Python/TS in v2 PRs) increases the need for parity docs and migration guides per language. **Feature requests aligned with actual usage** - Custom OpenAI endpoint URL (#6490) aligns with “provider swapping” reality. - Action/provider filtering + METATOOL/provider pattern discussion aligns with “too many plugins/actions” reality. --- ### 3) Implementation Opportunities (solutions per major pain point) #### Pain Point A (Docs): v2 onboarding, examples, and upgrade/migration clarity **High impact / low–medium difficulty** 1. **Single “Start Here (v2)” path**: one canonical doc landing page that answers: - “Use v2.0.0 branch, here’s why” - “Minimal working agent in 5 minutes” - “Database: pglite vs Neon, what changes, what doesn’t migrate” - “Where are Milady examples located?” - Similar approach: **Next.js “Getting Started” + “Upgrade Guide”** pattern (clear versioned docs + migration pages). 2. **Versioned migration matrix** (1.4 → 1.6 → 2.0) with explicit guardrails: - If auto-migration is removed, add a crisp “required stepping-stone” page + CLI messaging. - Similar approach: **Django/React** official upgrade notes with “breaking changes + required steps.” 3. **Reference “recipes” for popular stacks**: - “Agent + web server wrapper” - “Agent with image generation” - “Agent with Neon Postgres” - Similar approach: **LangChain cookbook** style recipes. **Medium impact / medium difficulty** 4. **Doc-driven discoverability of examples**: add an `examples/INDEX.md` that’s generated/validated in CI so “hidden examples in Milady” become searchable and stable. --- #### Pain Point B/F (Token & performance): context bloat, baseline overhead, duplicate calls **High impact / medium difficulty** 1. **Default context budget + provider/evaluator caps**: - Hard cap recentMessages/reflections length; add “summarize-to-fit” when exceeding budget. - Similar approach: **LlamaIndex context windows + summarization nodes**; many frameworks enforce truncation + summary memory. 2. **Action/provider gating by relevance by default**: - Productize ongoing work like action filtering (vector/BM25) into a stable default path for large plugin stacks. - Similar approach: tool routing/retrieval-based tool selection in **LangChain** and “tool choice” strategies in modern agent stacks. 3. **Fix cost-amplifying bugs first**: - Prioritize #6486 (URL → duplicate LLM calls). This is a direct 2x cost multiplier and visible UX defect. **Medium impact / higher difficulty** 4. **Provider METATOOL pattern**: - Investigate applying METATOOL pattern to providers (as discussed), but only after cheaper wins above. Gate behind a feature flag to avoid delaying releases. --- #### Pain Point C (Integration): plugin confusion + OpenAI-compatible endpoints + Ollama embedding failures **High impact / low difficulty** 1. **Deprecation + “preferred plugin” metadata**: - Mark `plugin-image-generation` as deprecated (if true) and point to `plugin-bootstrap` action in registry/docs. - Similar approach: **Homebrew formula deprecation warnings** / **npm deprecate** notices. 2. **Add `OPENAI_BASE_URL` (or equivalent) to OpenAI provider** (#6490): - Keep defaults unchanged; allow override for OpenAI-compatible vendors. - Similar approach: many OpenAI SDK wrappers support a `baseURL` override. 3. **Ollama Linux embedding troubleshooting guide + environment checks**: - Document known failure modes (model availability, CPU/GPU backend, library versions) and provide a diagnostic command. --- #### Pain Point D/E (Trust/Safety + Support UX): migration confusion, scams, ticketing **High impact / low–medium difficulty** 1. **One signed “Official Links” page** pinned everywhere (Discord + docs + GitHub): - Migration/status links, “we will never DM you,” ticket process, verification steps. - Similar approach: security posture pages used by major crypto projects during migrations/airdrops. 2. **In-Discord guided flow for migration/support**: - A bot command like `/migration-status` + `/support` that returns the canonical links and instructions. 3. **Clarify token naming and upgrade eligibility**: - Publish a short FAQ: “ElizaOS token vs milady.ai token (launch status), what upgrades apply to what.” --- ### 4) Communication Gaps (expectations vs reality) - **“v2 recommended” vs “v2 under-documented”**: users are instructed to adopt v2 but can’t find authoritative docs/examples, leading to repeated Qs (stability, HTTP wrapper, DB behavior). - **Migration/upgrade messaging ambiguity**: confusion over whether an “upgrade” applies to ElizaOS token or $milady.ai; users interpret ambiguity as risk. - **Roadmap visibility after restructuring**: core-dev announcement lists revenue projects but leaves unanswered questions (budgets per project, v2 importance, cloud ownership). This can erode contributor confidence if not addressed in a public roadmap note. **Documentation/onboarding gaps indicated by recurring questions** - “Which version is stable?” - “How do I use Neon vs pglite?” - “Where are the examples for the HTTP wrapper?” - “Which image generation plugin should I use?” - “How do I safely handle migration/tickets without scams?” **Targeted improvements** - Add an always-updated **“FAQ (top 10)”** in docs and pin it in Discord. - Add **CLI hints** for common misconfigurations (DB env vars, missing migration steps). --- ### 5) Community Engagement Insights **Power users / high-leverage contributors and their needs** - **Odilitime**: repeatedly surfaces performance/token issues (200k context), plugin guidance, and strategic direction—needs tighter performance tooling, profiling, and clearer “preferred patterns” to reduce support load. - **Stan ⚡**: deep DB refactor/migration work; suggests framework-level optimizations (MCP/METATOOL). Needs clear integration path (RFCs, ownership, merge windows). - **Ogie / buzzbysolcex**: shipping complex commercial plugins; needs stable registry processes, versioning guidance, and integration contracts. - **DarmaStef**: demonstrates real-world agent apps for non-technical users; needs templates, UI/examples, and deployment recipes to replicate. **Newcomer friction signals** - **BinaryCookies**: plugin choice uncertainty (image generation), version stability questions. - **KingRon / de_cryptkeeper / fanman22**: deployment/support and migration confusion; difficulty finding correct channels; high scam sensitivity. **Ways to convert passive users into contributors** - Create “good first issue” bundles around docs: - “Add recipe: Neon Postgres” - “Add recipe: image generation via plugin-bootstrap” - “Add troubleshooting: Ollama embeddings on Linux” - Monthly “office hours” for v2 onboarding + plugin best practices; capture Q&A into docs. --- ### 6) Feedback Collection Improvements **Current channel effectiveness** - Discord is capturing urgent confusion (migration/security, “which plugin?”) but is **unstructured**, leading to repeated questions and inconsistent answers. - GitHub issues capture actionable engineering items well (#6486, #6490), but many user confusions never become issues. **Improvements for structured, actionable feedback** 1. **Discord → GitHub issue intake bot**: - React-to-message to create a draft issue with template fields (version, runtime, repro steps). 2. **Standardized issue templates**: - Separate templates for “Provider compatibility,” “Token/cost regression,” “Migration/upgrade problem,” and “Docs missing.” 3. **Telemetry/diagnostics opt-in for token usage**: - A command that prints context composition (providers/evals/messages) so users can paste into issues. **Underrepresented segments** - Non-crypto “business automation” users (e.g., Gmail/WhatsApp/N8N) appear in repo progress but are not strongly present in Discord feedback this sample—likely missing UX feedback for integrations and onboarding. --- ## Prioritized High-Impact Actions (next 2–4 weeks) 1. **Fix #6486 (URL triggers duplicate LLM calls)** to eliminate 2x cost + duplicated output in webapp streaming. 2. **Publish a canonical v2 “Start Here” + migration matrix** (1.4→1.6→2.0, DB choices, example index, HTTP wrapper pointers). 3. **Add OpenAI provider `baseURL` override** (issue #6490) to unlock OpenAI-compatible endpoints with minimal surface-area change. 4. **Ship “Official Links + Anti-Scam + Migration FAQ” hub** and pin across Discord; add bot commands for `/support` and `/migration`. 5. **Implement default context budgets + summarization/truncation for recentMessages/reflections** (plus provider/evaluator caps) to reduce baseline token bloat for plugin-heavy agents.