bridle

An agentic harness for OpenAI-compatible chat completion endpoints.

Features

Harness user experience:

• Readline-based CLI with tab completion
  • Coloured output for sections (-Colour, or NO_COLOR env var)
• Oneshot mode (-p)
• Rewind turn history (/rewind and /turns)
• Support interrupting current message or tool call (^C)
• Session management (-Autosave, manual /save, and resume via -r)
• Import local files (wildcard match) into prompt (/read)
• Linux sandboxing with Bubblewrap (-Sandbox), use at own risk

Model integration with harness:

• Auto load AGENTS.md and CLAUDE.md files
• Compaction via context summarization (/compact)
  • Auto compact on client-side token threshold (-MaxContext) or on server-side length exceeded
• Tool calling
  • Available tools:
    • Local file I/O (read_file, write_file, apply_patch with automatic patch repair)
    • Command execution (local_shell)
    • Subagents (subagent with fork/fresh context modes)
    • Web search with DuckDuckGo
  • User confirmation levels (/checktools or -CheckTools: set to ask default, auto, or yolo)
    • Allow denying with alternative message (?)

Model settings:

• Use custom model provider (-Host and -Key)
• Toggle thinking (/think and /no_think)
• Customizable system message, default-tuned for Golang development

Usage

# For 128k 131072, for 256k 262144 . Mostly affects pp speed, not tg speed
llama-server -c 262144 --cache-type-k q8_0 --cache-type-v q8_0 -fa on -m /srv/llama/Qwen3.6-35B-A3B-UD-Q4_K_XL.gguf &

./bridle

# or
./bridle -CheckTools=yolo -Sandbox -Autosave=auto