Skip to content

Releases: blisspixel/deepr

v2.9.1 — Agentic Expert Chat, Skills System, deepr web

24 Mar 05:50
@blisspixel blisspixel
v2.9.1
4715021

Choose a tag to compare

View all tags
v2.9.1 — Agentic Expert Chat, Skills System, deepr web Latest
Latest

Highlights

Deepr v2.9.0–v2.9.1 introduces agentic expert chat and a domain-specific skills system, making experts interactive collaborators rather than just knowledge stores.

Agentic Expert Chat

  • Streaming chat over WebSocket with real-time token delivery, tool call visibility, and follow-up suggestions
  • 27 slash commands across 6 categories (Mode, Session, Reasoning, Control, Management, Utility)
  • 4 chat modes: ASK (quick answers), RESEARCH (default, all tools), ADVISE (structured recommendations), FOCUS (always-on chain-of-thought)
  • Visible reasoning: ThinkingPanel shows planning, search, evidence, and decision steps with confidence indicators
  • Human-in-the-loop approval with three tiers (auto-approve, notify, confirm) based on operation cost
  • Expert council: /council queries multiple experts in parallel, synthesizes agreements and disagreements
  • Task decomposition: /plan breaks complex queries into subtasks with dependency graph and parallel execution
  • Context compaction: /compact summarizes earlier messages to enable longer sessions
  • Memory commands: /remember, /forget, /memories for session-pinned facts
  • Conversations API for browsing and resuming past chat sessions
  • AI-generated expert portraits (SVG, cached per expert)

Expert Skills System

  • Domain-specific capability packages (skill.yaml + prompt.md + Python/MCP tools)
  • Three-tier storage: built-in, user global, expert-local
  • 4 built-in skills: web-search-enhanced, code-analysis, financial-data, data-visualization
  • CLI management: deepr skill list/install/remove/create/info
  • Tool namespacing and per-skill budget tracking
  • MCP tools: deepr_list_skills, deepr_install_skill

Expert Intelligence

  • Multi-provider consensus gap-filling (--consensus)
  • Semantic citation validation (--validate-citations)
  • Multi-pass gap-filling pipeline (--deep)
  • Automated gap discovery via claim clustering (deepr expert discover-gaps)
  • Conflict resolution with multi-provider adjudication (deepr expert resolve-conflicts)

v2.9.1 Fixes & Polish

  • deepr web CLI command to start the dashboard (--host, --port, --debug)
  • Fixed Windows file-locking race condition in provider metrics persistence
  • Added SECURITY.md and CODE_OF_CONDUCT.md
  • Consolidated pytest config into pyproject.toml

Stats

  • 4254 tests passing (0 failures)
  • 16 MCP tools, 31 API endpoints, 22 registry models across 5 providers
  • 12-page web dashboard with WebSocket push, skeleton loading, mobile nav

Full changelog: docs/CHANGELOG.md

Install:

pip install -e .          # Core
pip install -e ".[web]"   # With web dashboard
pip install -e ".[full]"  # Everything
Assets 2
Loading