2f675ee4bf
Cortex: FastAPI backend serving Inara via Claude/Gemini CLI backends. Includes SSE streaming chat, session persistence, Google Chat webhook handler, and Docker support. Inara: Identity files (persona, soul, protocols, memory, context tiers) mounted read-only into the container at runtime. Features in initial cut: - /chat endpoint with SSE keepalive + LLM fallback - Session store with rolling history window - Markdown rendering, copy-to-clipboard, links open in new tab - Stacked right-column input controls (height selector, enter toggle, note mode with public/private) — semi-hidden until textarea grows - /note endpoint for injecting public context into session history - Docker Compose config (local dev runs natively; Docker for server) Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
66 lines
1.8 KiB
Markdown
66 lines
1.8 KiB
Markdown
# CONTEXT_TIERS.md — Cortex Dispatcher Loading Spec
|
||
|
||
This file defines which Inara context files to inject into a session based on the target model's
|
||
context window. The dispatcher reads this to decide what to prepend.
|
||
|
||
---
|
||
|
||
## Tier 1 — Minimal (~1,500 tokens)
|
||
|
||
**Target:** Local models with ~8k context or less (Qwen 8B small, etc.)
|
||
|
||
**Load:**
|
||
- `SOUL.md`
|
||
- `IDENTITY.md`
|
||
- `USER.md` — first 30 lines only (identity + what he cares about)
|
||
|
||
**Notes:** Just enough for Inara to know who she is and who Scott is.
|
||
|
||
---
|
||
|
||
## Tier 2 — Standard (~5,000 tokens)
|
||
|
||
**Target:** Models with 16k–32k context (Haiku, Gemini Flash, Qwen 8B full)
|
||
|
||
**Load:**
|
||
- `SOUL.md`
|
||
- `IDENTITY.md`
|
||
- `USER.md` — full
|
||
- `MEMORY.md`
|
||
- `PROTOCOLS.md`
|
||
|
||
**Notes:** Full operational context. Sufficient for most routine tasks and conversations.
|
||
|
||
---
|
||
|
||
## Tier 3 — Extended (~15,000 tokens)
|
||
|
||
**Target:** Models with 32k–128k context (Sonnet, Gemini Pro, Qwen 14B, Qwen 30B)
|
||
|
||
**Load:**
|
||
- Everything in Tier 2
|
||
- `~/agents_sync/aether/docs/FLEET_MANIFEST.md`
|
||
- Most recent 2 session files from `sessions/`
|
||
- Relevant project doc (e.g., `CORTEX.md`) if task is project-related
|
||
|
||
---
|
||
|
||
## Tier 4 — Full (50,000+ tokens)
|
||
|
||
**Target:** Frontier models with 200k+ context (Claude Opus/Sonnet, Gemini 2.5 Pro)
|
||
|
||
**Load:**
|
||
- Everything in Tier 3
|
||
- Last 5–7 session files
|
||
- Full project docs as relevant
|
||
- `~/agents_sync/aether/docs/api_v3.md` if task involves Aether API
|
||
|
||
---
|
||
|
||
## Hard Rules
|
||
|
||
- `SOUL.md` and `IDENTITY.md` are **always** loaded, regardless of tier.
|
||
- **Never inject:** `.env` files, `TOOLS.md` (contains credentials), raw session logs older than 30 days.
|
||
- **MEMORY.md must stay under 4,000 tokens** — enforce this during distillation.
|
||
- When in doubt, use Tier 2. Over-loading small models degrades output quality.
|