Using prxy.monster with Aider

Aider is a terminal-based coding assistant that pairs with git. It supports custom OpenAI-compatible endpoints natively via CLI flags or environment variables.

Configure

Two ways — CLI flags or env vars. Pick one.

Option A: CLI flags

aider \
  --openai-api-base https://api.prxy.monster/v1 \
  --openai-api-key prxy_live_xxxxxxxxxxxxxxxxxxxxxxxx \
  --model gpt-4o

For Anthropic models:

aider \
  --anthropic-api-base https://api.prxy.monster \
  --anthropic-api-key prxy_live_xxxxxxxxxxxxxxxxxxxxxxxx \
  --model claude-sonnet-4-6

Option B: Env vars

export OPENAI_API_BASE=https://api.prxy.monster/v1
export OPENAI_API_KEY=prxy_live_xxxxxxxxxxxxxxxxxxxxxxxx
 
# OR for Anthropic
export ANTHROPIC_API_BASE=https://api.prxy.monster
export ANTHROPIC_API_KEY=prxy_live_xxxxxxxxxxxxxxxxxxxxxxxx
 
aider --model claude-sonnet-4-6

Option C: Aider config file

Put it in ~/.aider.conf.yml:

openai-api-base: https://api.prxy.monster/v1
openai-api-key: prxy_live_xxxxxxxxxxxxxxxxxxxxxxxx
model: gpt-4o

Code change

None — aider is a CLI; nothing to modify.

Verify

curl https://api.prxy.monster/health

Run aider and ask any question — successful response confirms routing.

What you get

Aider workflows are highly repetitive across a project — same files, same coding patterns, same fix categories. This is ideal for prxy.monster:

• Pattern memory — aider’s typical task (“fix this bug”, “add a test”, “refactor this function”) becomes more efficient as patterns accumulate.
• Semantic cache — re-asking similar questions across sessions returns cached answers.
• Cost guards — hard cap on per-day aider spend (aider sessions can rack up surprisingly).
• Infinite context — long aider sessions on big repos stop hitting the wall via ipc.

Recommended pipeline

PRXY_PIPE=semantic-cache,patterns,ipc,cost-guard

For high-volume aider users who want repeat-prompt savings:

PRXY_PIPE=exact-cache,semantic-cache,patterns,cost-guard

Streaming

Aider streams output by default. Pass-through.

Multi-model aider

Aider supports separate models for “main” and “weak” (autocomplete-style) tasks:

aider --model claude-sonnet-4-6 --weak-model claude-haiku-4-5 \
  --openai-api-base https://api.prxy.monster/v1 \
  --openai-api-key prxy_live_xxx

Both models route through prxy.monster.

Common issues

• --no-stream — works fine. Cache hits return as a single non-streamed response in this mode.
• Git integration — aider’s git operations are client-side. prxy.monster sees only the LLM calls.
• Context size errors — same as Cline; if aider’s diffs balloon the context, the ipc module helps but can’t compress an explicit user-supplied diff. Use a model with a large enough window.
• /voice mode (Whisper) — aider sends audio directly to OpenAI’s Whisper endpoint. This currently bypasses prxy.monster (no Whisper proxying yet).

Full example

Adapt examples/openai-quickstart — but for aider, the integration is purely CLI flags. No example app needed.