[WIP][18.0] [ADD] ai_connection#87
Closed
angelmoya wants to merge 3 commits into
Closed
Conversation
angelmoya
force-pushed
the
18.0-add-ai_connection
branch
2 times, most recently
from
10be16b to
d4f7c2e
Compare
…tence
This commit introduces a deep refactoring of the LLM execution core,
superseding previous implementations.
Key improvements:
- Unifies the core execution loop around a generator/iterator architecture.
- Introduces for robust tracking of conversational
state (draft, running, paused, pending_tool_approval, done, failed).
- Wraps tool execution in try/except blocks to return errors safely to the LLM
instead of crashing the Odoo transaction.
- Optimizes payloads by dynamically extracting base64 file data into
records.
- Adds an extensible hook to allow custom streaming
transports (e.g., temporary files) without blocking transactions.
- Introduces mode, allowing each iteration to be executed in an
isolated transaction, preventing long-running database locks.
- Introduces a nested model for deep JSON debugging of payloads, vacuumed by a cron.
- Extensive test coverage for persistence, streaming, tools, and attachments.
angelmoya
force-pushed
the
18.0-add-ai_connection
branch
from
d4f7c2e to
ef2c82a
Compare
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
3 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Supersedes: #77
📝 Context & Rationale
This Pull Request supersedes #77 . The original implementation for
AI tool execution was functional but suffered from structural limitations when dealing
with more complex AI interactions such as real-time streaming, human-in-the-loop (HITL)
validations, and robust conversational persistence.
Specifically, if a tool raised an exception, the entire Odoo transaction would crash.
Furthermore, continuous execution flows lacked the ability to pause and resume
effectively, and file attachments were bloat-heavy when kept purely in memory or JSON
structures.
This PR introduces a refactoring of the ai.connection core to solve these issues by
unifying the LLM execution flow around a generator-based architecture and providing
dedicated database models for execution persistence.
🛠️ Key Architectural Changes
• Unified Generator Core: _run and _run_ai have been refactored to consume a single
iterator logic ( handle_message_stream ). This streamlines both synchronous and iterative
workflows into a single manageable loop.
• Robust Persistence Layer: Introduced ai.connection.execution to track the state (
draft , running , paused , pending_tool_approval , done , failed ) and the
conversational history.
• Fail-Safe Tool Execution: Tool calls ( _execute_tool ) are now wrapped in a try/except
block. Instead of crashing the Odoo thread, exceptions are captured and returned to the
LLM as a JSON error payload, allowing the AI to naturally correct its inputs or notify the
user.
• Optimized Attachments: Base64 files inside prompts are now dynamically extracted into
ir.attachment records and rehydrated back only when communicating with the LLM API,
significantly reducing DB payload overhead.
• Extensible Streaming Architecture: While the exact streaming transport is not rigidly
defined out of the box, this refactor introduces an isolated _on_stream_batch hook. This
acts as a highly extensible foundation allowing developers to implement custom streaming
logic (for instance, streaming chunks via temporary files or websockets) without modifying
the core execution loop.
✨ New Features
• Backend UI for Executions: Admins can inspect active, paused, or failed executions from
the backend, viewing the full message history and token usage.
• Step-by-Step Mode (HITL & Transaction Isolation): Executions can now run in a stepwise
mode, pausing after each tool call to allow for manual inspection or future confirmation
dialogs ( pending_tool_approval ). Crucially, this allows each iteration to be executed in
a separate database query/transaction, preventing massive long-running transactions from
holding locks.
• Deep Debugging: Added a debug flag that records every single iteration in a nested
model ( ai.connection.execution.iteration ), dumping the exact request/response JSONs
exchanged with the API. A cron job automatically vacuums these debug records after 15 days.
I have also added a testing wizard ( ai.connection.run.wizard )!
You can now click the "Test Connection" button directly from the ai.connection form
view. It allows you to quickly test prompts, system prompts, and tools from the UI. You
can choose to run it in-memory to get a direct raw response, or store the execution in the
database to debug the step-by-step iterations. Test coverage is also included for it.
🧪 Testing
The test suite has been significantly expanded to cover:
• Standard and persistent executions.
• Paused states, stepwise increments, and interactive resumption.
• Stream chunk buffering logic.
• Automatic extraction and rehydration of file attachments from the JSON history.
• Tool error isolation and recovery.