Athena-Public

3. Install the SDK (optional — enables CLI commands)

---

# Lightweight install (~30 seconds, no ML dependencies)
pip install -e .

---

# Full install (~5–10 min, enables vector search and reranking)
pip install -e ".[full]"

⚠️ Don't pip install athena-cli — that's a different, unrelated package. Always install from inside the cloned repo.

💰 Cost

Athena is free. Forever. MIT licensed. You only pay for the AI subscription you're probably already paying for.

Try before you buy. Athena works with Google Antigravity's free tier — clone the repo, type /start, and see if it clicks. No credit card, no trial period, no catch. The free tier includes Pro models at reduced quota — enough to evaluate whether Athena works for you. Upgrade when you need higher throughput.

Why $200/mo sounds expensive — until you do the math. A single employee costs $5,000+/mo in salary, benefits, and management overhead. An AI agent on a max-tier subscription costs $200/mo, works 24/7, doesn't call in sick, and scales to any domain you throw at it. For best results, subscribe to the max plan on any one platform (Claude Max, Google AI Ultra, etc.) — the difference between $20/mo and $200/mo is the difference between a tool you use occasionally and a tool that runs your life. Heavy users routinely consume $2K–$3K+ in equivalent API costs per month — the flat subscription turns variable cost into fixed cost. For quota-limited plans (which is now the norm — Google has tightened rate limits significantly since April 2026), use /minmax mode to maximize output quality per token — the MinMax doctrine maintains the same quality floor while reducing token consumption by ~74%.

For peak performance, use /ultrastart every session. On a flat-rate plan, the marginal cost of deeper thinking is $0 — so the cost of under-thinking always exceeds the cost of over-thinking. /ultrastart loads ~20K tokens of structured context (identity, canonical memory, active state, semantic bridge) every session. On API pricing, this costs $2–5/session. On a $200/mo flat plan, it costs nothing. However: on quota-limited plans (most plans since Apr 2026), Athena defaults to /start (~10K tokens) with the MinMax protocol, reserving /ultrastart for high-complexity sessions. The pricing model of the underlying compute layer directly determines Athena's performance ceiling.

Boot cost is 2K–20K tokens (depending on mode) — constant whether it's session 1 or session 10,000. Details →

[!NOTE] Athena works with any model, but governance protocols and multi-step reasoning perform best with frontier models (e.g. Claude Fable 5, Gemini 3.5 Pro, GPT-5.6 Sol). Start with the free tier to test compatibility with your preferred model.

[!TIP] Save money getting started (updated Jul 2026). Google still offers a 1-month free trial on AI Pro for new subscribers, and AI Pro family sharing lets one subscriber extend AI benefits to up to 5 family members at no extra cost. Ultra pricing improved in May 2026: the top tier dropped from $249.99 to $200/mo, and a new $100/mo Ultra tier was added for lighter-but-regular development loads. Antigravity quotas were also raised roughly 9x from their April 2026 low point, so the old weekly-cap anxiety is mostly obsolete — though weekly limits still exist for sustained heavy use, and how Antigravity quota interacts with family sharing is not clearly documented (coordinate if mul