Agent fleets in three lines

Same code on your GPU, on Ollama, or on cloud APIs. Effort-based model routing, persistent memory and multi-agent coordination — without the lock-in.

Get started View source

bash

pip install awdk
adk quickstart
#   detects hardware, pulls models, configures backends
adk init my-agent && cd my-agent && python agent.py

3lines to start

3.8version

10packs

BSL 1.1license

Brain packs, ready to hand over

A pack bundles an identity, a system prompt, tools and skills into one unit an agent can be given wholesale. These ship with the package — each row opens the pack definition.

adk/packs10 of 10

Aither System OrchestratorAither — System Overseer & Orchestrator Brain Packaither · 2 skills Analyst StudioAnalyst — Data & Structured-ML Agent Brain Packanalyst · 2 skills BeadSpaceBeadSpace — an aither-adk agent pack for bead-spacebead-space Claude Code StudioClaude Code — Software Development Agent Brain Packclaude-code · 2 skills DGG ResearchDGG Research — brain packdgg_research · 1 skills GobboNet CompanionGobboNet Companion — an agent harness for a local-first chat clientgobbonet Hermes Architecture StudioHermes — Architecture & Reasoning Agent Brain Packhermes · 2 skills Iris Visual ArtisanIris — Visual Artisan Brain Packiris · 3 skills OpenClaw Research StudioOpenClaw — Web Research Agent Brain Packopenclaw · 2 skills PersonaPersona — an aither-adk agent pack for personapersona

Why it exists

Most agent frameworks bind you to one provider. This one treats the backend as configuration.

routing

Effort-based routing

Declare how hard a task is; the selector picks a model to match. Trivial work does not pay for a frontier model.

backends

Runtime switching

vLLM, Ollama and cloud APIs behind one interface. Change the backend without touching agent code.

memory

Knowledge-graph memory

Agents keep durable, queryable memory across runs rather than restarting cold every session.

fleets

Multi-agent coordination

Compose agents into fleets with shared context, so work can be decomposed and run in parallel.

packs

Tool packs

Bundle tools, prompts and identity into a pack — a reusable unit an agent can be given wholesale.

local

Local or cloud

No GPU? Set an API key. Have one? It auto-detects. Both? It routes by task complexity.

thinking

External thinking

Providers stopped returning chain-of-thought. Switch the native channel off, hand the model a scratchpad tool, and it reasons into the tool call — which comes back in plaintext.

infill

Fill-in-the-middle

Write the code BETWEEN two fragments, not around them. Ask a chat model to fill a gap and it rewrites your surrounding lines; this is the operation an editor completion actually needs.

Quick start

Install, set up inference, and run your first agent. The bootstrap installer handles Python for you if you would rather not manage it.

bash

# with Python already available
pip install awdk
adk quickstart
adk quickstart --cloud
adk start

# no Python? one line (macOS / Linux)
curl -fsSL https://aitherium.com/install.sh | sh

# Windows
powershell -ExecutionPolicy ByPass -c "irm https://aitherium.com/install.ps1 | iex"

Building something real? The Agent Dev Guide is the opinionated golden path — authoring a pack, never-forget RAG memory, BYO keys, and the gotchas worth skipping.

One estate, 8 doors

These projects are organs of the same system — the same identity plane, the same agent runtime, the same operational discipline. Follow any of them.

the osaitherium.comThe Living OS these four repos are cut from. Open, no login.pythonawdkBuild AI agent fleets — 3 lines, any backend, local or cloud.markdownawskillsPortable agent skills — self-contained procedures an agent loads on demand.markdownawknowledgeHow to run a coding agent so the result survives — the laws, with evidence.cudaaitherkvcacheNear-optimal KV cache quantization for LLM inference — sub-byte compression.powershellAitherZeroPowerShell 7+ automation framework — numbered, self-describing scripts.writingThe BlogHow all of it is actually built, measured and operated.pythonawrunA priority-aware queue and dispatcher for agentic runs and ad-hoc CI builds. It also judges whether the runner pool is big enough for the queue it is draining, and can ask a host to grow it -- reserving capacity is zero-sum, so a saturated pool needs more of it, not a different share of it.

The ecosystem this belongs to

Every Aitherium repository publishes an aither-manifest.json beside its page, and each surface reads all of them — so the network is browsable from any node in it. A repository that publishes nothing shows as unknown, never as zero.