Akmon is the open-source AI coding agent for regulated engineering. A single Rust binary that records every session as a tamper-evident, replayable, content-addressed artifact. Local-first, model-agnostic, built for teams that have to prove what their AI did. The only AI coding agent that turns each session into a portable evidence bundle, in an open format (AGEF), with deterministic replay and policy you can verify in CI.
Is this just observability? No. Akmon writes a tamper-evident audit chain plus a structured evidence summary per session. AGEF bundles are portable, content-addressed, and verifiable offline. Different audience, different shape.
How does this compare to Cursor, Claude Code, Copilot CLI? Those tools are general-purpose AI coding assistants optimized for productivity. Akmon is the one built for regulated codebases that need an audit chain, deterministic replay, and a portable evidence bundle by default.
How does this compare to Microsoft's Agent Governance Toolkit? That toolkit addresses runtime AI governance broadly. Akmon is the AI coding agent that produces evidence in AGEF, the open spec we maintain. Complementary, not competitive.
Does it work with my framework? Akmon is itself the agent. It supports MCP via --mcp-server, plus file ops, search, git, shell (policy constrained), and web fetch. Local and cloud providers are supported.
Report
Maker
Hey Product Hunt 👋
I have been building Akmon as the AI coding agent I would actually trust in a regulated codebase. v2.0.0 is the milestone where the trust pipeline (audit, evidence, SLO, replay) is real, AGEF is its own spec repo, and the tool surface is stable enough to put in CI. Apache-2.0, single Rust binary, no hosted runtime.
Try it in five minutes:
# 1. Install (example for Linux x86_64)
curl -L https://github.com/radotsvetkov/... -o ~/bin/akmon
chmod +x ~/bin/akmon
akmon --version
# 2. Set up a provider, pick one path.
# Path A, local-first with Ollama
ollama pull qwen2.5-coder:7b
# Path B, hosted, pick any one
export ANTHROPIC_API_KEY="YOUR_KEY"
# or: OPENAI_API_KEY / OPENROUTER_API_KEY / GROQ_API_KEY / Azure / Bedrock
# 3. Confirm provider health
akmon doctor providers
akmon config explain-provider
# 4. Run a small task
cd your-project
akmon --yes --output json --task "summarize failing tests and propose minimal fixes" | tee run.json
# 5. The trust pipeline
akmon audit verify .akmon/audit/<session-id>.jsonl
akmon evidence verify .akmon/evidence/<session-id>.json
akmon slo verify .akmon/evidence/<session-id>.json --strict
I am here all day to answer questions. The format is open, the spec is short, and the call for AGEF Bundle Profile implementations is open.
Report
No reviews yetBe the first to leave a review for Akmon
Frequently asked questions
Is this just observability? No. Akmon writes a tamper-evident audit chain plus a structured evidence summary per session. AGEF bundles are portable, content-addressed, and verifiable offline. Different audience, different shape.
How does this compare to Cursor, Claude Code, Copilot CLI? Those tools are general-purpose AI coding assistants optimized for productivity. Akmon is the one built for regulated codebases that need an audit chain, deterministic replay, and a portable evidence bundle by default.
How does this compare to Microsoft's Agent Governance Toolkit? That toolkit addresses runtime AI governance broadly. Akmon is the AI coding agent that produces evidence in AGEF, the open spec we maintain. Complementary, not competitive.
Does it work with my framework? Akmon is itself the agent. It supports MCP via --mcp-server, plus file ops, search, git, shell (policy constrained), and web fetch. Local and cloud providers are supported.
Hey Product Hunt 👋
I have been building Akmon as the AI coding agent I would actually trust in a regulated codebase. v2.0.0 is the milestone where the trust pipeline (audit, evidence, SLO, replay) is real, AGEF is its own spec repo, and the tool surface is stable enough to put in CI. Apache-2.0, single Rust binary, no hosted runtime.
Try it in five minutes:
I am here all day to answer questions. The format is open, the spec is short, and the call for AGEF Bundle Profile implementations is open.