AI-powered conflict resolution for Git
synthmerge is a minimalistic command-line tool that leverages AI to automatically resolve conflicts arising from Git commands. Built on the research of the Patchpal project, it provides a pure AI inference layer that seamlessly integrates with your existing Git workflow. While the AI generates code solutions, all code reviews and approvals remain within your favorite code editor.
-
Specialized AI Layer
Dedicated AI inference system that complements Git without duplicating its core functionality -
Git Integration
Leverages Git'sdiff3conflict markers as the foundation (requiresgit config merge.conflictStyle diff3) -
Editor Agnostic
Compatible with any development environment (VS Code, Emacs, Vim, etc.)
-
Universal Git Operation Support
Seamlessly integrates with all Git operations that create conflicts:cherry-pickmergerebaserevert
-
Model Flexibility
No fine-tuning required, any instruct large language model can be used -
Parallel Multi-AI Endpoint Support
Simultaneously queries multiple AI models to resolve conflicts:- Patchpal-backend (fine-tuned specifically for conflict resolution)
- Self-hosted open-weight open source LLMs with OpenAI-compatible endpoints
- Gemini (via OpenAI-compatible API)
-
Parameter Variants Support
Each AI endpoint can be configured with multiple parameter variants to run multiple inference strategies:- Different reasoning effort levels (high, medium, low)
- Temperature, top_p, top_k, min_p sampling parameters
- Context handling options (no_context flag)
-
Results Deduplication
Consolidates identical solutions and displays model and/or parameter variant agreement -
Review Using Your Workflow
- Resolved conflicts appear in your editor with model attribution
- AI-generated code requires manual review before commit
-
Fail-Safe Design
- When one model fails to resolve a conflict, Git's original conflict remains alongside solutions from other models for that hunk
- Each AI endpoint can be configured with timeout, delay, and max_delay parameters
-
Git sets up conflicts
git config merge.conflictStyle diff3 # Must be set git cherry-pick -x <commit> # Git detects conflicts
-
synthmerge analyzes conflicts
- Reads Git's
diff3conflict markers - Extracts context (3 lines before/after conflict)
- Generates precise AI prompt
- Reads Git's
-
AI resolves conflict
- Sends code + patch to configured endpoint
- Receives resolved code
-
Git gets updated
- synthmerge inserts the AI resolution into existing diff3 markers
- You review in your editor
✅ Works also for git rebase, revert and merge conflict resolutions.
Create ~/.config/synthmerge.yaml based on synthmerge.yaml:
endpoints:
- name: "Patchpal AI"
type: "patchpal"
url: "http://patchpal.usersys.redhat.com:9080/v1"
#timeout: 600
#retries: 10
#delay: 1000
#max_delay: 600000
- name: "llama.cpp vulkan simple"
url: "http://localhost:8811/v1/chat/completions"
type: "openai"
#no_context: false
- name: "llama.cpp vulkan"
url: "http://localhost:8811/v1/chat/completions"
type: "openai"
params:
# one query for each entry in the params list
- variant: "default"
- variant: "min_p"
temperature: 0.3
top_p: 1.0
top_k: 0
min_p: 0.9
- variant: "no_context"
no_context: true
- name: "Gemini 2.5 pro"
url: "https://generativelanguage.googleapis.com/v1beta/openai/chat/completions"
type: "openai"
model: "gemini-2.5-pro"
api_key_file: "~/.gemini.api-key"
params:
- reasoning_effort: "high"# Ensure Git is configured for diff3 conflict style
git config merge.conflictStyle diff3
# Attempt cherry-pick (will leave conflicts unresolved)
git cherry-pick -x <commit>
# Resolve conflicts with AI
synthmerge
# Review synthmerge resolved conflicts in each unmerged file ...
git diff --name-only --diff-filter=U
# ... or linearized in a single buffer to edit with ripgrep-edit
rg-edit -E vim -U -e '(?s)^<<<<<<<+ .*?^>>>>>>>+ '
rg-edit -E emacsclient -U -e '(?s)^<<<<<<<+ .*?^>>>>>>>+ '| Endpoint Type | Example Configuration | Notes |
|---|---|---|
| Patchpal-backend | type: "patchpal" |
Fine-tuned for patch resolution |
| OpenAI protocol | type: "openai" |
Self-hosted LLMs (e.g., llama.cpp) and Gemini |
✅ Gemini supports a compatible OpenAI endpoint
✅ Models work with stock weights – the prompt engineering simulates Patchpal's fine-tuned behavior.
The following statistics were generated using the synthmerge_bench tool on a C language dataset to evaluate model performance on conflict resolution tasks. These results may vary depending on prompt, context, and other variables.
Accuracy checks if the AI resolved conflict is an exact match including all spaces, tabs, and newlines.
Accuracy (aligned) checks equality of whitespace patterns up until the first non-whitespace character, ignoring differences in lines without non-whitespace characters and whitespace variations after the first non-whitespace character (i.e. Python equivalence).
Accuracy (stripped) compresses all whitespaces and newlines into a single space (i.e. C/C++/Rust/JavaScript equivalence).
The probability that at least one of the three differnt PatchPal beams is exact (not ignoring whitespace differences) is: 66.33% + 11.25% + 2.92% = 80.5%. This measurement used only new test data never exposed to the model during the fine tuning process.
# only the Beam 0 is comparable to the non Patchpal models
Model: Patchpal AI #0 (Beam search 0)
Accuracy: 63.33% (715/1129)
Accuracy (aligned): 67.76% (765/1129) # might be duplicate with other beams
Accuracy (stripped): 71.12% (803/1129) # might be duplicate with other beams
Error Rate: 0.53% (6/1129) # might be duplicate with other beams
# if Beam 0 is wrong, Beam 1 is right 11.25% of the time
Model: Patchpal AI #1 (Beam search 1) of the time
Accuracy: 11.25% (127/1129)
Accuracy (aligned): 22.50% (254/1129) # might be duplicate with other beams
Accuracy (stripped): 33.92% (383/1129) # might be duplicate with other beams
Error Rate: 0.53% (6/1129)
# if Beam 0 and Beam 1 are wrong, Beam 2 is right 2.92% of the time
Model: Patchpal AI #2 (Beam search 2)
Accuracy: 2.92% (33/1129)
Accuracy (aligned): 15.50% (175/1129) # might be duplicate with other beams
Accuracy (stripped): 25.86% (292/1129) # might be duplicate with other beams
Error Rate: 0.62% (7/1129)
Model: Gemini 2.5 pro (high) # reasoning_effort: high
Accuracy: 55.18% (623/1129)
Accuracy (aligned): 60.67% (685/1129)
Accuracy (stripped): 63.42% (716/1129)
Error Rate: 0.00% (0/1129)
Model: Gemini 2.5 pro (low) # reasoning_effort: low
Accuracy: 51.64% (583/1129)
Accuracy (aligned): 56.78% (641/1129)
Accuracy (stripped): 58.90% (665/1129)
Error Rate: 0.18% (2/1129)
# temperature: 0.7 top_k: 20 top_p: 0.8 min_p: 0
# llama.cpp vulkan Q6_K
Model: Qwen3-Coder-30B-A3B-Instruct (default)
Accuracy: 48.54% (548/1129)
Accuracy (aligned): 52.97% (598/1129)
Accuracy (stripped): 55.89% (631/1129)
Error Rate: 0.00% (0/1129)
# temperature: 0.7 top_k: 20 top_p: 0.8 min_p: 0
# llama.cpp vulkan Q6_K
Model: Qwen3-Coder-30B-A3B-Instruct (no_context)
Accuracy: 45.88% (518/1129)
Accuracy (aligned): 50.13% (566/1129)
Accuracy (stripped): 53.59% (605/1129)
Error Rate: 0.00% (0/1129)
Build from source:
git clone https://github.com/aarcange/synthmerge.git
cd synthmerge
cargo build --release
