aider

aider is the default coding client for AI sessions. It drives the model through the chat-completions API and applies edits as text — unified diffs or whole-file replacements — so it does not depend on the model's function-calling support, which is the least reliable part of serving a local model. Tool-calling agents such as opencode and Cline are covered on opencode and Cline; the shared session lifecycle (ai-session code, ai-session status, ai-session stop) is on Coding Sessions.

aider is already installed as part of the service; you do not install anything. It requires the working directory to be inside a git repository — run git init first if necessary.

Quick start

Step	Description	Command	Run on
1	Start a coding session (procedure on Coding Sessions)	`ai-session code`	Login node
2	Change into the git repository you want to edit	`cd /path/to/your/repo`	Login node
3	Run the aider command printed at start	See Step 1 below	Login node
4	Stop the session when finished	`ai-session stop`	Login node

A running session consumes SU whether or not you send requests

The default coding session (qwen2.5_coder_32B on 2 A100 GPUs) bills a floor of 2.0 SU per hour of wall-clock time; stop it with ai-session stop as soon as you stop working. Rates and the billing formula are on Billing and Service Units.

Step 1: Run aider

Run aider on the login node where you started the session, in a second terminal (leave the start terminal running). It can also run on your laptop through an SSH tunnel to the session's port; the tunnel procedure is on Coding Sessions.

ai-session code prints the exact command; the same command in terms of the session environment variables (set by eval "$(ai-session env)"; see Coding Sessions) is:

cd /path/to/your/repo
eval "$(ai-session env)"
OPENAI_API_BASE=$AISESSION_BASE_URL \
OPENAI_API_KEY=$AISESSION_API_KEY \
aider \
  --model openai/qwen2.5_coder_32B \
  --weak-model openai/qwen2.5_coder_32B \
  --model-metadata-file "$AISESSION_HOME/ai-session/aider_model_metadata.json" \
  --edit-format diff --analytics-disable

The printed command uses aider's full install path; after module load ai-session the plain aider above resolves the same way.
AISESSION_API_KEY is the session access key, minted at start. Every request must carry it; a request without it is refused with HTTP 401. See Coding Sessions for sharing it with your lab.
Replace /path/to/your/repo with the git repository you want to edit.

The two OPENAI_API_* variables are read by litellm, the client library aider uses to send requests in the standard OpenAI API format that most AI tools can talk to. The API base must include the /v1 suffix.

Flag	Purpose
`--model openai/<key>`	Selects the served model. The `openai/` prefix selects the standard format in litellm.
`--weak-model openai/<key>`	Routes aider's auxiliary requests (commit messages, history summarization) to the same local model rather than to `api.openai.com`.
`--model-metadata-file <path>`	Declares the model's context window (32768 tokens) and zero token cost to litellm. Without it, litellm cannot size prompts and prints `Unknown context window size`.
`--edit-format diff`	Requests unified-diff edits instead of full-file rewrites. Pass `--edit-format whole` instead if diffs are rejected for a given file.
`--analytics-disable`	Permanently disables aider's own usage telemetry. This is a client-side concern separate from the model traffic, which never leaves RCC; the Data location note on the home page covers the distinction.

The metadata file splits the 32768-token window as 28000 input tokens and 4096 output tokens, so prompt plus generated tokens cannot exceed the window. It carries entries for qwen2.5_coder_32B, qwen2.5_72B, qwen3_4b, and qwen3_32B; if you started the session with --model qwen2.5_72B, substitute openai/qwen2.5_72B in both --model and --weak-model.

Verification: aider starts its interactive prompt without printing Unknown context window size. At the prompt, type /tokens; it reports current context token usage against the window.

In-session commands

Command	Effect
`/add <path>`	Add a file to the editable context.
`/drop <path>`	Remove a file from the context.
`/ask <question>`	Ask a question without editing files.
`/tokens`	Report current context token usage.
`/clear`	Clear conversation history; added files remain.
`/run <cmd>`	Run a shell command and optionally add its output to the context.

aider maintains a repository map, so the model has structural awareness of files you have not explicitly added; add files with /add only when you intend to edit them.

Non-interactive use

To make a single edit and exit, for example from a batch script, add --yes-always (answers aider's confirmation prompts), --no-auto-commit (leaves the change uncommitted for review), and --message:

OPENAI_API_BASE=$AISESSION_BASE_URL \
OPENAI_API_KEY=$AISESSION_API_KEY \
aider \
  --model openai/qwen2.5_coder_32B \
  --weak-model openai/qwen2.5_coder_32B \
  --model-metadata-file "$AISESSION_HOME/ai-session/aider_model_metadata.json" \
  --edit-format diff --analytics-disable \
  --yes-always --no-auto-commit \
  --message "add type hints to the public functions in utils.py"

Verification: git diff in the repository shows the edit; nothing was committed.

Standard input can be piped in, for example to analyze a log file:

cat build.log | \
OPENAI_API_BASE=$AISESSION_BASE_URL OPENAI_API_KEY=$AISESSION_API_KEY \
aider --model openai/qwen2.5_coder_32B \
  --weak-model openai/qwen2.5_coder_32B \
  --model-metadata-file "$AISESSION_HOME/ai-session/aider_model_metadata.json" \
  --analytics-disable \
  --message "explain the traceback in this log and propose a fix"

Both examples assume eval "$(ai-session env)" has been run in the shell.

Errors

aider-specific error messages — Unknown context window size, the prompt reported as too long, and rejected diff edits — are listed with resolutions on Troubleshooting.