ourAgentTeams

My blog : https://aclitice.com

Run a local AI “team” from your terminal
中文 | English

---

What it is

ourAgentTeams is a command-line assistant for multi-step work: you describe what you want once; the app breaks it into pieces, routes each piece to a suitable model, tracks progress, and returns a single final answer — without you managing prompts for every sub-step.

The leader runs on your machine (via Ollama). You can stay fully local, or add cloud API models when you need stronger or specialized workers. Orchestration stays on your side; sensitive text is not sent to the cloud until you choose API workers (and even then optional redaction helps).

---

What you can use it for

• Software & writing — specs, refactors, docs, API design, boilerplate in chunks the system merges for you
• Research-style tasks — breakdowns, comparisons, structured reports (with optional retrieval from past runs)
• Day-to-day questions — interactive Single mode routes each message to a reasonable local model; Team mode is for bigger jobs you want planned and executed step-by-step
• Cost-aware workflows — mix free local models with paid APIs only where it matters; budget caps in config

---

What you get

• Interactive CLI — default session: chat with routing, or type /team … for full team planning + execution
• One-shot tasks — ouragentteams start "…" for non-interactive runs (scripts, CI, automation)
• Privacy helpers — scan for common secrets before sending text to external providers
• Model memory & reports — learn which models pay off; optional savings hints
• Optional tools — file / search / shell helpers in the workflow; RAG over past task text when enabled

---

Requirements

Ollama	Install and run on the same machine you use the CLI; pull at least the model set in config/config.yaml as leader.model
Python	3.11+

---

Install

Option A — script (picks venv / conda / uv when present)

bash setup.sh
# then: source .venv/bin/activate   # or follow the script’s hint

Option B — minimal

python3 -m venv .venv
source .venv/bin/activate
pip install -U pip
pip install -e .

Option C — see Makefile for conda / Docker targets.

Pull a model Ollama can run (match or update leader.model in config/config.yaml):

ollama pull qwen2.5:7b    # example; pick one that fits your GPU/RAM

---

How to use

After install, the ouragentteams command is on your PATH (same environment where you ran pip install -e .).

Step 1: Start interactive session

ouragentteams

• Single mode (default) — normal chat; leader routes each turn to a suitable model.
• Team mode — type /team <task> for plan + multi-worker execution.
• In-session helpers: /help, /mode, /clear, /exit.

Equivalent command:

ouragentteams chat

Step 2: Run one-shot tasks (non-interactive)

ouragentteams start "Your task in natural language"

Useful for scripts/automation or non-TTY environments.

Step 3: Manage Leader model

List local models (from Ollama):

ouragentteams leader list

Set Leader model and persist to config:

ouragentteams leader use qwen3.5:4b

Quick switch via option-style command:

ouragentteams leader switch --model qwen3.5:4b
ouragentteams leader switch --model qwen3.5:4b --persist

Step 4: Manage workers (CRUD-style)

List workers:

ouragentteams config list-workers

Add a local worker:

ouragentteams config add-worker --model gemma4:e2b --local

Add an API worker:

ouragentteams config add-worker --model gpt-4o --api-key <key> --strengths "coding,analysis"

Remove a worker:

ouragentteams config remove-worker --model gpt-4o

Update worker config: remove then add again with new fields (there is no dedicated update command yet).

Verify connectivity for leader + all workers:

ouragentteams config verify

Step 5: Useful ops

ouragentteams reload    # reload config.yaml
ouragentteams report    # model performance summary
ouragentteams --help    # all commands

---

Project layout (quick map)

Path	Purpose
config/config.yaml	Leader model, workers, budget, privacy
data/	Sessions, history, vector store (created at runtime)
DEVELOPER_GUIDE.md	Deep docs for developers

---

Tests

pip install pytest
pytest -q

---

License

MIT — see the LICENSE file in the repository root.