atlas/secondbrain
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity Actions

add Components-Roadmap

atlas 2026-07-01 19:28:15 +00:00
parent dc250f5eba
commit cffa9d73b2
1 changed files with 46 additions and 0 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

46
Components-Roadmap.-.md Normal file

@ -0,0 +1,46 @@
# Components Roadmap
SecondBrain is the first component extracted from the Atlas local-first autonomy stack into
its own clean, generalized, secret-free repo. This page tracks the **sister components** worth
the same treatment — the result of a full read-only survey of the stack.
Extraction rules (applied to SecondBrain, applied to all):
1. **No secrets, ever** — config moves to env / example files; hardcoded creds are scrubbed and the git history starts clean.
2. **Generalize** — strip host/path/tenant specifics; sane local-first defaults.
3. **Redeployable** — README + `.env.example` + install + systemd/Docker.
## Ship-ready (config already externalized — low effort)
| Component | Proposed repo | What it is | Effort |
|---|---|---|---|
| **LLM router** | `llm-mux` | OpenAI-compatible `/v1/chat/completions` gateway that routes across Anthropic + local/remote Ollama nodes, with a cheapest-that-fits auto-router, circuit breakers, and streaming. Single FastAPI file, fully env-driven. | LOW |
| **Notifier** | `notify-mux` | Zero-dependency fan-out of title/body/priority to self-hosted ntfy push + WhatsApp (Evolution). Config in an external JSON. | LOW |
| **Approval gate** ⭐ | `agent-gate` | A human-in-the-loop "brake": money / external-comms / irreversible / self-modify actions are parked in a SQLite approvals queue pending release; reversible actions pass and log. The liability shield that makes agent autonomy safe to resell. | MED |
| **Content-aware monitor** | `site-sentinel` | Uptime monitor that checks *body content* (size + marker), not just HTTP 200, and self-heals by restoring from a mirror + restarting the service. | MED |
## Needs generalization first (valuable, more coupled)
| Component | Proposed repo | Note |
|---|---|---|
| Multi-node Ollama orchestrator + file job-queue | `ollama-cluster` | Node table → config; pairs well with `llm-mux`. |
| Odoo invoice CLI (create → PDF email → cloud upload) | `odoo-invoice-cli` | Odoo-schema-specific; config already in env. |
| Stock-driven newsletter (HTML + PDF) builder | `stock-newsletter` | ⚠️ must scrub a hardcoded mail credential before extraction. |
## Keep internal (Atlas-specific glue)
The perception→cognition→action "Mind" spine (`atlas_loop`, `atlas_orchestrator`, the `sense_*`
collectors, effectors) is tightly bound to the private event kernel and tenant workflows. The one
cleanly-extractable keystone from it is the **approval gate** above.
## Do NOT create original repos (upstream forks only)
Deployments of third-party OSS (LiteLLM, Glance, n8n, Firefly, Paperless, Nextcloud, MeshCentral,
Open WebUI, Stalwart, Authentik, Forgejo, …) are config around upstream projects — fork upstream
only if a custom patch is needed.
## Security gate on every extraction
Two internal tools carry hardcoded credentials and must be scrubbed **before** any extraction:
a newsletter mailer and a document-bridge API token. The router, notifier, gate, invoice, and
sentinel candidates are already secret-free in source (config externalized) — those are the
cleanest to ship first.
---
_Recommended order:_ `llm-mux` → `notify-mux` → `agent-gate` → `site-sentinel` → the rest.