Teams Auto-Response
Paste a Teams thread, draft a thorough IT reply with triage → writer → polish, copy rich text straight back into Teams.
Bring-your-own-model · air-gap ready
Route work across local multi-model pipelines or cloud APIs when you choose. Ground answers in RAG, MCP, or files you drop yourself. Ship with security controls and MDM-ready JSON — without surrendering your data by default.
Konoha AI — Part of the AI Suite — michaelnancarrow.com, burnerchat.org, openmodelresponder.org
Platform
OpenModelResponder is local-first infrastructure for AI workloads — not a hosted chat that monetises your prompts. Bring your own models. Decide when (if ever) traffic leaves.
Run entirely offline with local models. Optional API / cloud routing is an explicit decision — never a silent dependency. Ideal for regulated floors, OT islands, and teams that cannot send corporate content to a third-party SaaS by default.
Assign roles across local models (triage, writer, polish, embed) or route selected stages to cloud / multi-model APIs when policy allows.
Ground generation in retrieval over dropped files, knowledge folders, and MCP-connected tools — so answers reflect your environment, not a generic internet guess.
Token-gated UI, owner-only secrets, cloud consent gates, and withhold rails so internal docs / skills / soul stay local when a remote model is in the path.
Configurational JSON you can ship, version, and manage like any other endpoint policy — tone, depth, models, paths, and feature toggles without rewriting the binary.
Bring-your-own-model via Ollama today; size candidates with llmfit; pull what fits. No vendor lock on a single hosted weight.
The UI states whether you are local-first with API support, or actively sending content off-machine — so operators are never surprised.
Use cases
Teams Auto-Response is a flagship workflow — not the only one. Same local stack powers knowledge, automation, design, and everyday chat.
Paste a Teams thread, draft a thorough IT reply with triage → writer → polish, copy rich text straight back into Teams.
Turn resolved incidents and runbooks into structured articles — keep them in your local KB / doc/ store for the next retrieval.
Assist with scripts, pipelines, and operator automation on the same machine — without shipping proprietary code to a public cloud chat.
Iterate copy, layout notes, and component ideas against local models — useful when brand or customer data must stay inside the perimeter.
A private assistant for questions, summaries, and exploration — still under your routing and security controls.
Operator path
One use case, end to end — the same controls apply to other workloads.
Run on loopback with your models (python3 respond.py --serve). Air-gap friendly; no account required to think locally.
Drop files into knowledge folders, wire MCP where available, or rely on skill packs — retrieval beats hallucination for environment-specific work.
Keep stages local, or approve cloud / API models per run. Security controls withhold sensitive pack and doc bodies from remote paths.
Copy a Teams-ready reply, publish a KB draft, or feed automation — config JSON keeps the behaviour consistent across managed devices.
Routing
Decide where each hop runs. Default is on-box; remote is consent-gated.
| Path | What runs | Leaves the machine? |
|---|---|---|
| Local multi-model | BYOM roles — triage / writer / polish / embed | No (optional public reference fetches only) |
| Cloud / API hop | Selected roles via ollama.com or future OpenRouter / HF | Yes for that hop — docs / skills / soul / agent bodies withheld |
| Air-gapped mode | Local models only; cloud disabled in config / MDM JSON | Nothing leaves by policy |
Security
Wired-in rails — not a brochure checkbox.
Click around
One featured use case in chrome. Generate is simulated — no models, no egress.
| Role | Model |
|---|---|
| triage | llama3.2:3b |
| writer | gemma4:latest |
| writer_expanded | qwen3.6:latest |
| polish | llama3.1:8b |
| embed | nomic-embed-text |
| Model | Size | Compute | Assigned |
|---|---|---|---|
| gemma4:latest | 5.2 GB | fast — fits in RAM easily | writer |
| llama3.2:3b | 2.0 GB | fast — fits in RAM easily | triage |
| nomic-embed-text | 0.3 GB | RAG embed · local only | embed |
allow_cloud=false · webui_host=127.0.0.1 · withhold packs on remote · skills_paths=["skills"]
Soul (local): “Prefer numbered steps. Never invent ticket numbers.”
FAQ
Straight answers. Expand any item.
No. It is a static product story + educational mock. The live tool is local software with your models and config — not this website.
Yes — that is a primary design goal. Keep cloud disabled, use BYOM local weights, and ground work in files / RAG you control.
No. It is featured because it is concrete and proven in the codebase. The same platform story covers KB generation, local automation, web design assistance, and general private chat.
Platform direction: ground context via RAG / MCP / dropped files, and manage behaviour through configurational JSON suitable for MDM. See the live repo for what ships today versus what is on the roadmap.
No. Generate shows fake progress and a canned reply. Nothing leaves your browser for inference.
From the repo: cd projects/Invoke-TeamsResponse, copy config.example.json → config.json, then python3 respond.py --serve.