- (llm) Classify transient service-unavailable backend errors
- (llm) Skip transient-unavailable backends with per-backend cooldown
- (pipeline) Hold vision-incomplete messages pending, never false-success
- (resume) Re-OCR on retry; exhaustion is failure, never reported as success
- (llm) Extract reusable CircuitBreaker from ollama
- (llm) Rename is_service_unavailable to is_service_available
- (config) Document circuit_open_secs on cloud backends + vision ollama
- Use coarser Duration units flagged by clippy after the MSRV bump
- (pipeline) Cover vision-unavailable short-circuit and resume re-OCR guard
- (deps) Raise MSRV to 1.95 so
rand = "0"resolves to rand 0.10 - (release) 0.3.2
- (deps) Pin rand to 0.10 to stop resolver picking rand 0.8
- (llm) Fix stuck enrichment, harden backends, split LLM modules
- (error) Drop unwrap/expect on infallible and local paths
- (test) Split test modules over 500 lines
- (llm) Dedup forced-summary, fix panics, share retry backoff
- (free_router) Defer startup pool fetch off multi-thread runtime
- (llm) Cover chain helpers, memory tools, and tool dispatch
- (telegram) Test the api url before connection
- (release) 0.3.0
- Support preprocessing
- Semaphore ollama calls
- (tg) Accept the media groups
- (tools) Support duckduckgo search
- Add web links to ROAM_REFS if used
- (llm) Thinking tool
- (llm) Memory support
- (memory) Migrate from sqlite to grafeo
- Improve metrics
- Add feedback support
- Enforce LLM memory usage
- (message) Add serde derives for pending store serialization
- (pending) Implement SQLite pending store with migrations
- (render+config) Add inbox_pending tag and ResumeConfig
- (output) Add atomic org-file entry patcher
- (adapters) Add Telegram resume notifier and status msg tracking
- (resume) Add background resume task with pending queue metrics
- (pipeline+main) Integrate pending store and resume task
- (llm) Add free_router backend for dynamic free-model pool
- (llm) Record enrichment metadata in org entries
- Prevent the networking issues
- (web) Update the urls to prevent 404
- (tg) Apply timeouts for the first reply
- (tg) Support timeouts on attachment download
- (tg) Enforce ipv4
- (output) Use correct ids for nodes and attachment paths
- (llm) Enforce chain and ensure the result is reported
- (output) Correct org-mode formatting
- (reconnect) Use biased select to fix flaky shutdown-before-first-run test
- (memory) Suppress noisy vector index warning on startup
- (llm+resume) Add connect_timeout and handle exhausted pending items
- (llm) Improve Ollama stability with circuit breaker and robust JSON parsing
- (llm) Redact api_key from tracing spans in call_chat_completion
- (memory) Retry Grafeo open when DB is briefly locked
- (adapters) Extract reconnection loop
- (CLAUDE.md) Forbid unwrap/expect/panic in production code
- Refresh CLAUDE.md and README.md for current project state
- Update the attachment test
- Improve coverage
- Add coverage for pending store, resume, render, and pipeline
- Add targeted unit tests to lift coverage in low-cost modules
- Raise coverage for resume_task and http adapter
- Cover telegram resume notifier and web proxy/ui paths
- Cover pipeline::process and url_fetcher download/rewrite paths
- Update Cargo.toml docs
- (web) Refurbish UI
- Code style cleanup
- (llm) Prevent OOM
- Style cleanup
- Colorize errors
- Add SQL workflow to CLAUDE.md and sqlfluff config
- Document [pipeline.resume] in config.example.toml
- Bump deps
- (lint) Clear pre-existing clippy warnings and enforce crate-wide deny
- Initial commit
- Support crawl4ai
- (web) Support logs in ui
- (ollama) Support thinking/not thinking modes
- (http) Extract the data from html
- (tool) Support retries
- (tools) Support Kagi search
- (telegram) Accept attachments
- Add status reporting
- Update the content extraction
- (telegram) Reconnect on net problems
- (http) Support redirects and compression
- (llm) Add debug prints for llm requests
- (http) Use built-in CA
- (http) Support CORS
- (telegram) Prevent dispatcher panic on start
- (pipeline) Do not enrich the enriched article
- Support LLM prompts configuration
- Improve coverage
- Improve coverage
- Cleanup
- Anodize more functions
- Bump deps
- (http) Support http retries
- Sync up with the config
- Cleanup docs and build