Use Aider with GammaInfra

Point Aider at GammaInfra's smart router and you get dollar-cost visibility per edit (not just token counts), the cost-quality dial for cheap-mode vs quality-mode per session, and automatic provider fallback when one provider rate-limits in the middle of a multi-file refactor. Setup is two env vars — Aider's existing CLI works unchanged.

The pain Aider users hit

Aider is precise — every edit is a diff, every diff is reviewable, every refactor goes through git. That precision has a cost shape Aider users know well:

• Per-edit cost grows with repo size. Aider sends a chunk of relevant repo context with every call. Heavy edits on big codebases can cost 5–10× what you'd expect from the prompt alone.
• Token counts ≠ dollars. Aider prints token counts after each edit but you have to translate that into provider rates yourself.
• Rate-limit walls mid-session. Aider sessions naturally fan out — 20+ edits in a refactor. OpenAI/Anthropic throttle, you wait.

What changes with GammaInfra

Point Aider at GammaInfra's smart router with two env-var changes — its existing CLI works unchanged on top of the OpenAI SDK shape. After the swap:

• Every response carries X-GammaInfra-Cost-USD. The dashboard at dashboard.gammainfra.com rolls these up so you can see exact spend per session.
• Provider rate-limits cascade. When Anthropic throttles your key, GammaInfra's fallback chain serves the next provider (Google, then DeepSeek, etc.) — Aider sees a 200 and moves on.
• You can switch quality tiers per command via headers without re-launching Aider. X-GammaInfra-Cost-Quality: 0.7 for cheap refactors, 0.0 for hard architectural changes.

Setup

1. Get a GammaInfra API key

Sign up at gammainfra.com and verify your email. $3 trial credit covers maybe 30–50 Aider edits on a small repo, plenty to feel the difference.

2. Set environment variables

The cleanest way — put these in your shell rc file (.bashrc, .zshrc, etc.):

export OPENAI_API_BASE="https://api.gammainfra.com/v1"
export OPENAI_API_KEY="sk-gammainfra-..."

Aider reads these environment variables automatically — no flag needed.

3. Or pass them as CLI flags

aider --openai-api-base https://api.gammainfra.com/v1 \
      --openai-api-key sk-gammainfra-... \
      --model gammainfra/auto

4. Or put them in .aider.conf.yml

If you want repo-local config (or want different settings per repo):

# .aider.conf.yml (in your repo root)
openai-api-base: https://api.gammainfra.com/v1
openai-api-key: sk-gammainfra-...
model: gammainfra/auto

5. Pick a model

Aider benefits from quality models for harder edits. Some patterns:

• gammainfra/auto — task-aware routing. Reasonable default for most repos.
• anthropic/claude-opus-4-7 — Anthropic flagship. Best for complex refactors and tricky type-system work.
• anthropic/claude-sonnet-4-6 — Sonnet. Balanced quality + cost.
• openai/gpt-5 — OpenAI flagship.
• deepseek/deepseek-v4-pro — Cheap reasoning tier with thinking mode. Slower per-edit but materially cheaper on big sessions.
• gammainfra/cheap — cost-optimized routing. Use for docs edits, comment fixes, simple type adjustments.

Cost-conscious patterns

• Cost-quality dial per session. Pass --extra-headers 'X-GammaInfra-Cost-Quality: 0.7' when you start aider for a heavy refactor that doesn't need top-tier reasoning. Switch to 0.0 for the gnarly stuff.
• Max-latency budget. X-GammaInfra-Max-Latency-Ms: 30000 as a header — GammaInfra returns 504 if any single edit takes more than 30 seconds, instead of Aider quietly stalling on a slow provider.
• BYOK for high-volume. If you already pay OpenAI/Anthropic directly, add those keys via the dashboard's Provider Keys tab. GammaInfra charges 1–2% per request instead of the pass-through model.

Verify it's working

Start an aider session, make one edit, then in the dashboard:

• Recent requests show with model, provider, latency, exact cost.
• Aider's own cost line (after each commit) reflects actual provider rates — no markup.
• If you ever care about the underlying provider chosen, the X-GammaInfra-Endpoint response header reveals it.

Trade-offs

• Latency. ~10–50 ms overhead per call. Imperceptible during the slower think-then-diff cycles Aider does anyway.
• Cost. 3% top-up fee (launch window) / 5% (standard) on managed credits. Pass-through token rates — no markup. Compares favorably with Aider users currently topping up OpenAI/Anthropic credits directly.
• Privacy. Prompts and responses aren't logged by default. See privacy policy.

Troubleshoot

• Aider says "model not found". Your --model argument is malformed. Use slash-namespaced format: gammainfra/auto, not auto. Or use provider direct-pins: anthropic/claude-opus-4-7.
• Tool calls fail on some models. Aider uses tool calls for architect mode and structured edits. Pin a high-quality model (claude-opus-4-7) if you hit tool-call failures with gammainfra/auto.
• Token counts look off. Aider's count includes its internal prompt; the cost-USD header reflects actual provider tokens including system prompts the provider adds.

Ready to try it?

Get a GammaInfra API key →

$3 free trial credit on signup, $10 minimum top-up. Pass-through provider token rates plus 3% top-up fee during the launch window (5% after 2026-06-23).