Stack
other
Kind
enhancement
Description
Runnable integration demos for common agent frameworks and plain Python loops — same session export shape across stacks.
Deliver minimal runnable integrations (or documented stubs where deps heavy) under integrations/<stack>/. Do not add one folder per model brand — group by how the body is wired.
| Path |
Body / models |
AURA wiring |
Status |
integrations/skillware/ |
Skillware skills (+ optional Ollama body) |
SkillwareHost reference |
shipped |
integrations/openai/ |
OpenAI API (GPT, …) |
body + Skillware egress |
shipped |
integrations/anthropic/ |
Anthropic API (Claude) |
body + Skillware egress |
shipped |
integrations/google/ |
Google Generative AI (Gemini) |
body + Skillware egress |
shipped |
integrations/ollama/ |
Local Ollama HTTP body — swap OLLAMA_MODEL for Llama, Qwen, DeepSeek, Hermes, etc. |
with ag.session() + model.call on spine |
open (#20 / PR #53) |
integrations/openai-compat/ (or README section) |
Any OpenAI-compatible HTTP API — DeepSeek, Moonshot/Kimi, OpenRouter, vLLM, local gateways |
same loop shape as openai/; base_url + key from .env |
open (doc + thin example) |
integrations/google/adk/ (stub) |
Google ADK agent runtime |
wrap ADK run; AURA session around tool/agent steps |
open (stub OK if deps heavy) |
integrations/langgraph/ |
LangGraph graph tools |
wrap graph invoke; egress on tool nodes |
open (stub in #19) |
integrations/crewai/ |
CrewAI crew kickoff |
session wraps crew execution |
open |
integrations/openai-agents/ |
OpenAI Agents SDK (distinct from chat completions loop) |
session + tool egress adapter |
open |
examples/08-emit-only-loop/ |
plain Python |
manual emit only |
shipped |
examples/ + ToolHost |
mock / Skillware host |
ToolHost / SkillwareHost |
shipped (05–07) |
Not separate integrations: model names alone (Hermes, Llama 3, Qwen 2.5) — use Ollama tags or an OpenAI-compat endpoint. Claude and Gemini stay under Anthropic and Google rows above.
Paths, .env vars, or run command (optional)
Each integration README: .env vars, run command, export shape (JSONL + summary + audit report), Audit posture profile, link to onboarding (#13).
Out of scope: full framework forks; auto-discovery (roadmap) — manual wrap only.
Stack
other
Kind
enhancement
Description
Runnable integration demos for common agent frameworks and plain Python loops — same session export shape across stacks.
Deliver minimal runnable integrations (or documented stubs where deps heavy) under
integrations/<stack>/. Do not add one folder per model brand — group by how the body is wired.integrations/skillware/SkillwareHostreferenceintegrations/openai/integrations/anthropic/integrations/google/integrations/ollama/OLLAMA_MODELfor Llama, Qwen, DeepSeek, Hermes, etc.with ag.session()+model.callon spineintegrations/openai-compat/(or README section)openai/;base_url+ key from.envintegrations/google/adk/(stub)integrations/langgraph/integrations/crewai/integrations/openai-agents/examples/08-emit-only-loop/emitonlyexamples/+ ToolHostToolHost/SkillwareHostNot separate integrations: model names alone (Hermes, Llama 3, Qwen 2.5) — use Ollama tags or an OpenAI-compat endpoint. Claude and Gemini stay under Anthropic and Google rows above.
Paths, .env vars, or run command (optional)
Each integration README:
.envvars, run command, export shape (JSONL + summary + audit report), Audit posture profile, link to onboarding (#13).Out of scope: full framework forks; auto-discovery (roadmap) — manual wrap only.