🦄 evi-run — Customizable Multi-Agent AI System v1.1.1

Fixing the 'private' mode error.

🦄 evi-run — Customizable Multi-Agent AI System v1.1.0

Highlights

• Separated DEX analytics into a dedicated agent, improving tool routing clarity and modularity.
• Fixed web search routing and several edge cases in tool selection.
• Improved answer rendering stability and safety (HTML splitting, code blocks, comment escaping).
• Updated localization (EN/RU) and README.

What’s Changed

Features / Improvements

• agents_: enhance tool routing; add explicit policies and wording; integrate dedicated DEX Research agent; strengthen scheduler integration
  • Commit: 7feec6f — “Fixed a web search issue. DEX analytics is separated into a separate agent. Some other tool routing errors have been fixed.”
  • Commit: 60ce657 — “send_answer: dynamic budget, safe HTML splitting, pre/code fixes, HTML comments escaping; agents_: tool routing wording; cleanup (remove dead code)”
  • Commit: c09d043 — “update instructions in bot/agents_tools/agents_.py”

Fixes

• Web search routing fixes and guardrails for tool selection
  • Commit: 7feec6f

Localization

• Update localization files (I18N) — new/updated strings for help and scheduler, EN/RU parity improvements
  • Commit: 5f3a6a7 — “Update localization files (I18N)”

Documentation

• README: Python version badge, quick edits and clarifications
  • Commits: 8307425, 90759d1, 94bdbb1, a4c0dce (PGP-signed)

Commits

• 7feec6f — Fixed a web search issue. DEX analytics is separated into a separate agent. Some other tool routing errors have been fixed.
• 94bdbb1 — update README.md
• c09d043 — update instructions in bot/agents_tools/agents_.py
• 90759d1 — update README.md
• 31083a9 — update README.md
• 60ce657 — send_answer: dynamic budget, safe HTML splitting, pre/code fixes, HTML comments escaping; agents_: tool routing wording; cleanup (remove dead code)
• 65b2381 — chore(gitignore): ignore tests/, pytest artifacts and local pytest.ini
• 8307425 — Update Python version in README
• b4084f5 — Merge branch 'main' of https://github.com/pipedude/evi-run
• a4c0dce — Update README.md (PGP-signed)

---

If you find issues, please open an issue on GitHub or reach out via our Telegram community: https://t.me/evi_run

evi-run v1.0.1

Key Changes

• Runtime time (UTC) injected into the main agent’s system instructions.
• Date handling: forbidden to use dates from training data; “now/today” is taken from the Runtime Context.
• Tool routing: deep_knowledge is not used for operational data/news; for facts — use Web Search first.

Details

• File: bot/agents_tools/agents_.py
  • Added datetime, computing now_utc (ISO8601, UTC).
  • Main agent instructions switched to an f-string with a RUNTIME CONTEXT (do not ignore) block.
  • Updated TOOL ROUTING POLICY: current time comes from the Runtime Context (not via Web Search).
  • Strengthened EXECUTION DISCIPLINE: must not infer “today” from cutoff.
  • Updated deep_knowledge: analysis relative to Runtime Context; up-to-date facts via Web Search.

Fixed

• Agent no longer inserts an “old” date from training data in responses or tool invocations.

Compatibility

• Breaking changes: none. Only instructions and the source of “current time” were changed.

Changelog

• feat(agent): inject runtime UTC context; fix date handling policies; adjust deep_knowledge routing (bot/agents_tools/agents_.py)

😈 evi-run — Customizable Multi-Agent AI System v1.0.0

We're excited to announce the first major release of evi-run! This release introduces powerful new capabilities including intelligent task scheduling, enhanced AI agents, and improved system architecture.

✨ Major Features

🤖 AI Task Scheduler

• Smart task planning - Schedule tasks with natural language requests
• Multiple scheduling types:
  • One-time tasks with specific date/time
  • Daily recurring tasks
  • Interval-based tasks (every X minutes)
• Full CRUD operations - Create, read, update, delete tasks
• Timezone awareness - Automatic conversion to UTC for scheduling
• Persistent storage - Tasks survive system restarts

👥 Multiple Admins Support

• Flexible admin management - Support for multiple administrators
• Enhanced permissions - Admin-only commands and elevated privileges
• Proper credit allocation - Different limits for admins vs users
• Secure access control - Admin verification throughout the system

🧠 GPT-5 Integration

• Advanced reasoning - Deep Agent now powered by GPT-5
• Structured analysis - Multi-step logical reasoning capabilities
• Enhanced research - More accurate and comprehensive investigations
• Configurable verbosity - Adjustable detail levels in responses

🛠️ Technical Improvements

Architecture Enhancements

• Modular design - Better separation of concerns
• Persistent scheduler - SQLAlchemy-based job storage
• Performance optimizations - MCP server connection caching
• Improved error handling - Better resilience and debugging

Database Improvements

• New UserTasks model - Complete task management schema
• Enhanced repositories - CRUD operations for tasks
• Fixed credit logic - Proper admin vs user credit allocation
• Migration support - Seamless database updates

🌐 User Experience

Enhanced Localization

• Updated capability descriptions - Clear, structured feature lists
• New translations - Task scheduler messages in EN/RU
• Performance tips - User guidance for optimal token usage
• Improved onboarding - Better first-time user experience

AI Personality Improvements

• Clearer communication - Evi explains capabilities more effectively
• Better task routing - Smarter delegation to specialized agents
• Enhanced memory - Improved context retention and recall

🐛 Bug Fixes

• Fixed admin credit allocation logic
• Improved error handling in task execution
• Enhanced MCP server stability
• Better timezone handling for scheduled tasks
• Fixed various edge cases in multi-agent routing

---

Full Changelog: a87460e...4201acf

Installation: See README.md for setup instructions

Documentation: Updated with new features and configuration options

Agents config overhaul (adding GPT‑5, tool routing, instructions)

Highlights

• Major updates in [bot/agents_tools/agents_.py]
  • Main agent: expanded instructions (persona, communication style, execution discipline, date‑handling policy) and stricter anti‑hallucination guidance.
  • GPT‑5 family readiness: main agent prepared for GPT‑5 (currently gpt‑4.1), deep agent prepared for GPT‑5 (currently o4‑mini); examples include model_settings (Reasoning/Verbosity).
  • Tool routing: clarified call policies and ordering across conversation memory, knowledge base, and web search.

Docs

• README: updated examples to gpt‑5/gpt‑5‑mini with model_settings.

Chores

• .gitattributes: enforce LF line endings at the repo level.

Compare

v0.9.1...v0.9.2

v0.9.1 - Fix for daily credits allocation

Bug Fixes

• Fixed a bug in the daily credits allocation system for users

Technical Details

• Corrected logic in database/repositories/utils.py for proper calculation and allocation of daily credits

evi.run v0.9.0 - Initial Release

This is the first public beta release of evi.run, a multi-agent AI system with Telegram integration.

✅ Implemented Features

• 🧠 Memory Management and Context Handling
• 📊 Document Processing (PDF, images, text, Markdown, Word, PowerPoint, Python files)
• 🌐 Web Intelligence and Smart Search
• 🔍 Deep Research
• 📈 DEX Analytics
• 💱 Token Trading
• 🎨 AI Image Generation
• 🤖 Custom Agent Creation

🚧 Features in Development

• Pay Mode: Token-based monetization system (works in test mode, available upon request)

Available Usage Modes

• ✅ Private Mode (fully implemented)
• ✅ Free Mode (fully implemented)
• 🚧 Pay Mode (partially implemented)

Requirements

• Ubuntu 22.04 server
• Telegram Bot Token, OpenAI API Key
• Minimum: 2 CPU cores, 2GB RAM, 10GB storage

For detailed installation and usage instructions, please refer to the README.md