Release v2.8

Release v2.8 — conversational memory, one-click pivots, reasoning-model fix

Merges feature/v2.8:

• Session-scoped follow-up chat grounded in each investigation
• One-click suggested pivots that launch a new investigation
• Fix blank Findings for reasoning models (o1 / DeepSeek-R1) — #137
• Strict-grounding rule on summary presets
• Investigation output reformatted as Markdown headings + bullets
• Prompt-template brace-injection hardening

Release v2.7

fix: resolve config shadowing crash for custom/LM Studio models

In resolve_model_config(), the local variable 'config' shadowed the
imported config module. For models not in _llm_config_map (LM Studio,
custom API), the local was None, causing AttributeError on
config.CUSTOM_API_MODEL. Rename local to 'cfg' to fix the crash.

Release v2.6

docs: update README for v2.6 (OpenAI-compatible API, Custom API Provider sidebar)

llama.cpp fix, generic OpenAI-compatible API, Ollama Docker hint, provider prefix dropdown

• Fix llama.cpp/LM Studio thinking models (Qwen3.5, Gemma 4): auto-append /v1 to
  LLAMA_CPP_BASE_URL in resolve_model_config, set streaming=False to avoid
  'No generations found in stream' from reasoning_content chunks

• Add localhost bypass in _ensure_credentials so local model users aren't
  incorrectly prompted for OPENAI_API_KEY

• Add generic OpenAI-compatible provider via CUSTOM_API_BASE_URL + CUSTOM_API_KEY:
  fetch_custom_api_models() auto-discovers models, resolve_model_config routes
  them through ChatOpenAI with the custom base URL

• Log actionable Docker hint when Ollama is unreachable at localhost/127.0.0.1
  (use host.docker.internal instead)

• Add get_model_display_names() and format_func to model selectbox so the
  dropdown shows [provider] model_name labels without changing underlying keys

Release v2.5

chore: CI run for recent merged PRs (#116, #118)

PR #116 — Harden scraping pipeline (Chintanpatel24)

• Thread-local reusable HTTP sessions with retry + connection pooling
• Response size limit (1MB cap) to prevent memory exhaustion
• Content-type allowlist — only text/HTML responses parsed
• Worker concurrency clamped to safe range

PR #118 — Config, error UX, and scraper polish (Chintanpatel24)

• _clean_env(): strips whitespace and accidental quotes from .env values
• OpenRouter 401 error hints: specific guidance for invalid/expired keys
• st.stop() when no models are configured — prevents silent continuation
• scrape.py: deduplication, streaming download, URL scheme validation, debug logging

Release v2.4

Add UI enhancements: organized sections, save/load investigations, result limits

• Replace stat cards with Notes/Sources/Findings sections after pipeline run
• Sources expander shows all filtered .onion links with titles
• Notes expander shows refined query, model, domain, and result counts
• Findings section replaces Investigation Summary heading
• Auto-save completed investigations as JSON to investigations/ directory
• Past Investigations sidebar panel to load and re-display saved runs
• Max Results to Filter slider (10-100, default 50) caps results before LLM filter
• Max Pages to Scrape slider (3-20, default 10) caps filtered results before scraping
• README: add tip for mounting investigations/ volume in Docker

Co-Authored-By: Claude Sonnet 4.6 noreply@anthropic.com

Release v2.3

Add custom prompts, domain presets, and provider config check

• Add PRESET_PROMPTS dict to llm.py with 4 research domain variants:
  threat intel, ransomware/malware, personal identity, corporate espionage
• Update generate_summary() to accept preset and custom_instructions params;
  custom instructions append to the base prompt (never replace)
• Add Prompt Settings sidebar expander with domain preset selector,
  live system prompt viewer (read-only), and custom instructions textarea
  with domain-specific placeholder examples
• Add Provider Configuration sidebar section that auto-checks all 6
  LLM providers (OpenAI, Anthropic, Google, OpenRouter, Ollama, llama.cpp)
  on page load with clear set/missing/optional status indicators
• Show prominent error banner when no models are available due to missing
  env vars, surfacing the root cause of issue #99

Co-Authored-By: Claude Sonnet 4.5 noreply@anthropic.com

Release v2.2

Remove CLI mode, make Robin UI-only

Robin now runs exclusively through the Streamlit web UI. The Click-based CLI
router (main.py) is removed, Docker entrypoint launches Streamlit directly,
and the binary build workflow is replaced with a release-only workflow.

• Delete main.py (Click CLI router)
• Update entrypoint.sh to exec streamlit run ui.py directly
• Update Dockerfile to expose port 8501
• Remove yaspin (CLI-only dep) from requirements.txt
• Replace binary.yml with release.yml (keep versioning, remove PyInstaller build)
• Update README to reflect UI-only usage

Co-Authored-By: Claude Opus 4.6 noreply@anthropic.com

Release v2.1

Add health check feature - LLM API and search engine connectivity checks in UI

Co-Authored-By: Claude Opus 4.6 noreply@anthropic.com

Release v2.0

Major Updates - added new openrouter models, llama.cpp support, and improved error handling - version bump to v2

Release v1.15

Added 2 new Tor Search Engines