agentforge

Build entire organizations of AI agents — roles, personalities, experience, skills, and tools — then run them, or export them to the agent framework of your choice and sync them to your project-management suite. Fully interoperable with the Cognis 300+ suite (edgemesh / fleet / jtf-meridian) and the popular agent + PM platforms.

What is this?

Most agent tools make you wire up one or two bots in code. agentforge lets you design a whole org — like staffing a company — and treat it as data: a Research Firm, a Software Team, or Cognis's own JTF MERIDIAN command structure, each with named agents that have a job title, a personality (a careful analyst, a relentless skeptic, a pragmatic builder…), an experience level, the skills they're good at, and the tools they can use. Once you've described the org, agentforge can run it on a task (locally, or on your own AI fleet), or hand it to whatever agent framework you already use — CrewAI, AutoGen, LangGraph, OpenHands, TaskWeaver, MetaGPT — and put its tasks on your board in GitHub, Jira, Linear, Notion, Trello, or Asana. One definition of your agent workforce; many places to run it.

The model

Organization → Teams → Agents. Each Agent has a Persona (archetype + traits like rigor/creativity/risk/skepticism + tone), an experience level (intern…principal…exec), skills (from a registry), tools (mapped to real integrations and Cognis capabilities), goals, a model hint, and a reports_to. Everything round-trips to JSON/YAML.

What it does

agentforge templates                       # ready-made orgs
agentforge org --template jtf_meridian     # the Cognis command structure as agents
agentforge agents --template research_firm # roles / personas / skills / tools
agentforge run --template software_team --team eng --task "design an API"   # +--fleet for live
agentforge export --template research_firm --framework crewai               # or autogen/langgraph/openhands/taskweaver/metagpt
agentforge export --template software_team --framework autogen --code       # runnable snippet
agentforge pm --template software_team --platform github --task "ship A"    # or jira/linear/notion/trello/asana
agentforge frameworks                       # everything it interoperates with

Interoperability

Agent frameworks (export): CrewAI · AutoGen/AG2 · LangGraph · OpenHands · TaskWeaver · MetaGPT (+ DeepTeam red-team persona mapping). One org definition → each framework's native config (and runnable code for the code-first ones).

Project-management suites (sync): GitHub Projects/Issues · Jira · Linear · Notion · Trello · Asana — agents become assignees, tasks become issues/cards.

Cognis suite: edgemesh is the model/compute backend (the FleetBackend targets its OpenAI-compatible /v1); the local fleet/cog4 runs each agent (per-agent slot via Agent.model); jtf-meridian ships as a built-in org template; the tool registry maps to kycaml / cryptotrace / ragkit / labforge; labforge plans the hardware the org runs on. Run agentforge frameworks.

Runtime

Runtime(backend).run_team(org, team, task) assigns a task across a team, collects each agent's contribution, and has the lead synthesize a result, with per-agent error isolation and usage metering. LocalMockBackend is deterministic/offline; OpenAIBackend targets any OpenAI-compatible endpoint (retries + timeout); FleetBackend targets edgemesh/the fleet with a graceful fallback.

Enterprise

Built to drop into a real org (see docs/ENTERPRISE.md):

• Bring your own backend via env vars (AGENTFORGE_BACKEND/_BASE_URL/_API_KEY/_MODEL) — OpenAI, Azure OpenAI, vLLM, a gateway, or the local fleet; secrets from your store, never code.
• Governance — policy as code: global + per-role tool allow/deny lists, approval-required tools, budget caps, chain-of-command. PolicyEngine.validate(org) gates deploys; agentforge validate runs it in CI.
• Audit trail: append-only JSONL of every agent run/synthesis/error with timestamps.
• Org-as-config in version control (load_org/save_org), reviewed and validated like code.
• Reliability: retries/timeouts, per-agent error isolation, usage metering for cost/limit tracking.
• Runs on infrastructure you own — pair with edgemesh + labforge for on-prem/air-gapped.

agentforge validate --file orgs/eng.json --require-reports-to --approval-required github,sql

Install

curl -fsSL https://raw.githubusercontent.com/cognis-digital/agentforge/HEAD/install.sh | sh   # Linux/macOS

irm https://raw.githubusercontent.com/cognis-digital/agentforge/HEAD/install.ps1 | iex        # Windows

pipx install "git+https://github.com/cognis-digital/agentforge.git"   # or uv tool install / pip install
git clone https://github.com/cognis-digital/agentforge.git && cd agentforge && pip install .  # source

Then: agentforge --help

Topics / Domains

ai-agents · multi-agent · agent-orchestration · llm · crewai · autogen · langgraph · agentops · part of the Cognis Neural Suite (AI & ML domain).

Verification

tests   : 10 passing (deterministic; offline mock backend)
runtime : pure Python standard library; no third-party deps

Disclaimer

agentforge is a definition + orchestration + export layer; it does not vendor or reimplement the target frameworks — install those you actually run. Exporters produce each framework's config/structure; PM adapters produce payloads (wire a client + token to push). Not affiliated with any framework or platform named.

License

Cognis Open Collaboration License (COCL) 1.0 — see LICENSE.