How long does the integration take?

Change one base URL in an OpenAI-compatible client, or install the Python or TypeScript SDK; the quickstart routes a first request in a few minutes.

Will this lock us into another vendor?

No. The router is Apache-2.0 open source, your provider keys stay yours, and the API is OpenAI-compatible, so leaving is a change of base URL.

What happens when a provider fails?

The router falls back to the next model your policy allows and records the fallback in the trace, so an outage costs a retry, not an incident.

audit - Prompt - OpenSmartRoute

Name: audit
Author: Ahmad-al-Rifai

Prompt file imported from Ahmad-al-Rifai/rag-engine (.claude/commands/audit.md). Copyright stays with the author.

Run the Phase One audit of AIOS. Read-only: do not modify files, git state, packages, models, or ChromaDB data.

Read CLAUDE.md and docs/HANDOFF_CONTEXT.md first. Treat the handoff context as claims to verify.

1. Environment

Determine from evidence: OS (Windows / WSL / Linux), shell, Python version, dependency manager, and any existing venv. Do not print env var values.

2. Git

Report the repo root, remotes (strip any credentials from URLs), current branch, git status, and the last ~10 commits. Do not fetch or pull. Say whether local-vs-GitHub sync can be determined without fetching.

3. Codebase map

Find what actually exists for each component: entry points · FastAPI routes · Markdown discovery · parsing/chunking · metadata · embeddings (which model?) · ChromaDB init/persistence path/collection names · retrieval · prompt building · LLM provider(s) · config handling · dependencies · tests · docs.

Draw the real data flow as a short text diagram. Mark missing components as missing.

4. Safe runtime checks

Run only checks that need no installs, downloads, or writes:

Does the venv import the app?
Run existing tests if present.
Is the configured vault path readable? (Report counts only; don't read note contents.)
Can ChromaDB be opened read-only and its collection count reported?
Is Ollama reachable (/api/tags), and is mistral present?
If the API can start without side effects, hit /health.

For each check you skipped, say why.

5. Report

Use these sections:

Verified facts
Historical claims: confirmed / contradicted / unverified
Inferences (labelled as such)
Unknowns, noting which ones only Ahmad can answer
Working / broken / missing, mapped to Milestones 1–7 in CLAUDE.md
Biggest blocker
Recommended next step (one), then a short prioritized plan (≤5 items)

Then propose concrete values for the ## Commands section of CLAUDE.md and wait for approval before editing anything.

Install audit and it becomes one more candidate the router can pick - when it fits.

A listing is a routing target with a manifest: what it does, which domains it covers, what it costs and who publishes it. Once installed it sits beside your own models and tools, is scored like any other candidate for each request, and shows up in the trace when it wins. Ratings come from workspaces that installed it, one per account.

Run the Phase One audit of AIOS. **Read-only: do not modify files, git state, packages, models, or ChromaDB data.** Read `CLAUDE.md` and `docs/HANDOFF_CONTEXT.md` first. Treat the handoff context as claims to verify. ## 1. Environment Determine from evidence: OS (Windows / WSL / Linux), shell, Python version, dependency manager, and any existing venv. Do not print env var values. ## 2. Git Report the repo root, remotes (strip any credentials from URLs), current branch, `git status`, and the last ~10 commits. Do **not** fetch or pull. Say whether local-vs-GitHub sync can be determined without fetching. ## 3. Codebase map Find what actually exists for each component: entry points · FastAPI routes · Markdown discovery · parsing/chunking · metadata · embeddings (which model?) · ChromaDB init/persistence path/collection names · retrieval · prompt building · LLM provider(s) · config handling · dependencies · tests · docs. Draw the real data flow as a short text diagram. Mark missing components as missing. ## 4. Safe runtime checks Run only checks that need no installs, downloads, or writes: - Does the venv import the app? - Run existing tests if present. - Is the configured vault path readable? (Report counts only; don't read note contents.) - Can ChromaDB be opened **read-only** and its collection count reported? - Is Ollama reachable (`/api/tags`), and is `mistral` present? - If the API can start without side effects, hit `/health`. For each check you skipped, say why. ## 5. Report Use these sections: - **Verified facts** - **Historical claims**: confirmed / contradicted / unverified - **Inferences** (labelled as such) - **Unknowns**, noting which ones only Ahmad can answer - **Working / broken / missing**, mapped to Milestones 1–7 in CLAUDE.md - **Biggest blocker** - **Recommended next step** (one), then a short prioritized plan (≤5 items) Then propose concrete values for the `## Commands` section of CLAUDE.md and wait for approval before editing anything.

{ "prompt": "Run the Phase One audit of AIOS. **Read-only: do not modify files, git state, packages, models, or ChromaDB data.**\n\nRead `CLAUDE.md` and `docs/HANDOFF_CONTEXT.md` first. Treat the handoff context as claims to verify.\n\n## 1. Environment\nDetermine from evidence: OS (Windows / WSL / Linux), shell, Python version, dependency manager, and any existing venv. Do not print env var values.\n\n## 2. Git\nReport the repo root, remotes (strip any credentials from URLs), current branch, `git status`, and the last ~10 commits. Do **not** fetch or pull. Say whether local-vs-GitHub sync can be determined without fetching.\n\n## 3. Codebase map\nFind what actually exists for each component:\nentry points · FastAPI routes · Markdown discovery · parsing/chunking · metadata · embeddings (which model?) · ChromaDB init/persistence path/collection names · retrieval · prompt building · LLM provider(s) · config handling · dependencies · tests · docs.\n\nDraw the real data flow as a short text diagram. Mark missing components as missing.\n\n## 4. Safe runtime checks\nRun only checks that need no installs, downloads, or writes:\n- Does the venv import the app?\n- Run existing tests if present.\n- Is the configured vault path readable? (Report counts only; don't read note contents.)\n- Can ChromaDB be opened **read-only** and its collection count reported?\n- Is Ollama reachable (`/api/tags`), and is `mistral` present?\n- If the API can start without side effects, hit `/health`.\n\nFor each check you skipped, say why.\n\n## 5. Report\nUse these sections:\n- **Verified facts**\n- **Historical claims**: confirmed / contradicted / unverified\n- **Inferences** (labelled as such)\n- **Unknowns**, noting which ones only Ahmad can answer\n- **Working / broken / missing**, mapped to Milestones 1–7 in CLAUDE.md\n- **Biggest blocker**\n- **Recommended next step** (one), then a short prioritized plan (≤5 items)\n\nThen propose concrete values for the `## Commands` section of CLAUDE.md and wait for approval before editing anything.", "variables": [], "notes": "Read-only audit of the AIOS repo, environment, and runtime. No changes.", "metadata": { "source": { "provider": "github-claude-commands", "repository": "https://github.com/Ahmad-al-Rifai/rag-engine", "path": ".claude/commands/audit.md", "ref": "23905e7a7d5a14c729669987cabf550f6d0878f5", "url": "https://github.com/Ahmad-al-Rifai/rag-engine/blob/23905e7a7d5a14c729669987cabf550f6d0878f5/.claude/commands/audit.md", "key": "Ahmad-al-Rifai/rag-engine/.claude/commands/audit.md" } } }

audit

About

1. Environment

2. Git

3. Codebase map

4. Safe runtime checks

5. Report

Install audit and it becomes one more candidate the router can pick - when it fits.

What it costs to adopt - in engineering time.

One base URL

No vendor lock-in

Failover built in

Use it

Manifest

Reviews