# antoinezambelli/forge

A Python framework for self-hosted LLM tool-calling and multi-step agentic workflows

Repository: https://github.com/antoinezambelli/forge
Canonical: https://ross.abutalabs.com/products/antoinezambelli-forge
Language: Python
License: MIT
License Family: permissive
Topics: agentic-ai, agentic-workflow, agents, function-calling, llama-cpp, llamafile, llm, ollama, python, self-hosted, tool-calling
Last push: 2026-08-26T20:53:16+00:00

## Health v2 (maintenance only)
Score: 82/100 (v2, computed 2026-09-03T02:20:16.233290+00:00)
- activity 99, release rhythm 99, longevity 14
- inputs: {"age_days": 198, "days_push": 7, "days_rel": 7, "gap_med": 6, "n_releases_24m": 16}
- flags: none
- formula: round(0.45*activity + 0.35*rhythm + 0.20*longevity); archived -> min(score, 10)

## Adoption (not part of the score)
Stars 2227, forks 174 (observed 2026-08-28T04:06:28.789107+00:00)

## What it is
Forge is a Python framework providing a reliability layer for self-hosted LLM tool-calling, with guardrails like rescue parsing, retry nudges, and response validation. It can be used as a drop-in proxy server (OpenAI/Anthropic compatible), a WorkflowRunner for structured agent loops, or composable middleware inside your own orchestration loop.

## Use cases
- make local 8B models reliable at tool calling
- run multi-step agentic workflows with self-hosted LLMs
- proxy Claude Code or aider through guardrails to a local model
- rescue malformed tool calls from small LLMs
- share one GPU inference slot across multiple agent workflows
- add guardrails to an existing agent orchestration loop
- connect OpenAI-compatible coding tools to Ollama or llama.cpp

## When to choose
- you self-host models via Ollama, llama.cpp, Llamafile, or vLLM and need dependable tool calling
- you want a transparent proxy that upgrades existing coding agents without rewrites
- you need structured, constrained agent loops with required steps and validation
- you want composable reliability middleware rather than a full agent orchestrator

## When to avoid
- you need multi-agent graphs, DAG planning, or cross-agent coordination
- you only use hosted APIs and don't care about local model reliability
- you want a full coding agent rather than a layer beneath one
- your project requires Python below 3.12

## Facets
- artifact type: framework
- maturity: active
- function: agent-framework, llm-inference, middleware, http-server, sdk
- domain: artificial-intelligence, large-language-models, self-hosted, developer-tools
- platform: python, self-hosted, cli, cross-platform
- tags: tool-calling, guardrails, proxy-server, ollama, llama-cpp, vllm, function-calling, agentic-workflows, local-llm, ai-agents

## Member repositories
- antoinezambelli/forge (main) score 82

## Provenance
- Observed fields: from GitHub, fetched 2026-08-28T04:06:28.789107+00:00.
- Health v2: computed from the inputs above; adoption is never an input.
- Inferred fields (summary, facets, guidance): AI-extracted, prompt v1, taxonomy v1, on 2026-08-30T02:44:49.207422+00:00, confidence not recorded.
  - readme: https://github.com/antoinezambelli/forge (fetched 2026-08-28T04:06:28.789107+00:00, sha 4528460b2023)
- Data as of 2026-08-30T08:39:29.467469+00:00.