Fetching metadata from the HF Docker repository...
-
Skill: quote only from opened pages, not search snippets; validator reads Reddit post bodies and skips logged queries
-
Stricter quoting and counting rules in the study skill; report validator script
-
Headless runs end only on an explicit [[FINAL]] marker instead of phrase matching
-
reddit_search tool via the Arctic Shift archive (reddit.com blocks bots)
-
OpenResearcher tier (small/auto), 'study' depth, final answer at the step limit, surface tool errors
-
Make the fast model configurable (DEXTER_FAST_MODEL); gpt-5.4-mini is not served by freellmapi
-
Headless runs continue when the model narrates a plan or asks how to proceed
-
Headless queries finish the task instead of asking to continue
-
Recover tool calls lost in streaming; keep slow query streams alive
-
Keyless web search (SearXNG + Jina), OpenResearcher tool, llama provider, empty-reply retry, public-opinion-study skill
-
Rate-limit model failover (switch mid-session to next in fallback chain, with cooldown) on both stream + invoke paths
-
Model selector (default minimax-m2.7, native tool-calling) + generous retries/patience (6 attempts, 180s timeout) (part 2)
-
Model selector (default minimax-m2.7, native tool-calling) + generous retries/patience (6 attempts, 180s timeout)
-
Raise depth caps (quick8/balanced12/deep20) so data queries complete despite redundant tool calls
-
Adaptations: web channel profile, depth control, markdown+skills+depth in Dexter, Ask-Dexter context passing, Research widgets board, RULES.md (part 2)
-
Adaptations: web channel profile, depth control, markdown+skills+depth in Dexter, Ask-Dexter context passing, Research widgets board, RULES.md
-
Fix dexter SSE disconnects: Bun idleTimeout 255s + proxy swallows mid-stream cutoffs (part 2)
-
Fix dexter SSE disconnects: Bun idleTimeout 255s + proxy swallows mid-stream cutoffs
-
Add flow zoom +/-/fit controls; harden proxy against transient upstream disconnects
-
LLM: freellmapi via openai provider (custom base URL), helmholtz stays Blablador
-
Fix: hedge board height (h-screen) + agent default model env override (part 2)
-
Fix: hedge board height (h-screen) + agent default model env override
-
Unified Dexter/HedgeFund/OpenBB app (FastAPI single-server, no Caddy/terminal) (part 3)
-
Unified Dexter/HedgeFund/OpenBB app (FastAPI single-server, no Caddy/terminal) (part 2)
-
Unified Dexter/HedgeFund/OpenBB app (FastAPI single-server, no Caddy/terminal)
-
initial commit
