✦        ✦        ✦

           ▓▓▓
           ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓
         ▒▒▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▒▒
         ▒▒▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▒▒
           ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓
             ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓
               ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓
                   ▓▓▓▓▓▓▓▓▓▓▓▓
                    ▓▓      ▓▓
                    ▓▓      ▓▓
                 ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓
               ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓
             ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓
           ▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓▓

AKMON

AI coding sessions you can verify, replay, and defend in review.

Akmon is a terminal-first AI coding agent for regulated engineering teams that need explicit policy boundaries, auditable tool activity, and portable session evidence across local development and CI.

regulated engineering focus tamper-evident evidence single rust binary policy boundaries replay + diff + verify provider optionality
↓ Install Read the docs → v2.0.0 release notes → GitHub
akmon — ~/my-project — claude-haiku — Anthropic
~/payments-service $ akmon
▓▓▓▓ akmon v2.0.0 · ~/my-project · claude-haiku · Anthropic

You: trace token validation flow and list policy-relevant risks

→ semantic_search "token validation flow"
✓ semantic_search 4 results [Tab to expand]
→ read_file src/auth/middleware.rs
✓ read_file 287 lines
→ read_file src/api/token_guard.rs
✓ read_file 164 lines

Akmon: Validation happens in src/auth/middleware.rs and src/api/token_guard.rs.
I found expiry + issuer checks, and one endpoint missing audience validation.

a1b2c3d4 · tokens:4,821 · cache:8,779 · ~$0.004 · step 3
Live preview
// unique to akmon

See the agent loop

Realistic workflow preview for regulated teams: scoped discovery, policy-gated changes, then verifiable evidence checks. The sequence below cycles through a representative run.

akmon — live session preview

Step 1 — You steer; Akmon proposes tools.

// why akmon

Built for review,
not just generation

Akmon is designed for teams that need defensible software change history: explicit permissions, portable session evidence, deterministic checks, and provider control.

📋

Session evidence first

Every run can produce structured audit and evidence artifacts that survive beyond chat history and support review workflows.

🔒

Policy boundaries

Filesystem, shell, network, and tool actions are policy-evaluated, so sensitive operations are governed instead of implied.

⚙️

Single Rust binary

Terminal-first operation in local shells, SSH, and CI without dependency on editor plugin runtime stacks.

🔑

Provider optionality

Use local or hosted models with your own credentials and routing choices, instead of binding engineering workflows to one vendor.

💰

Cost transparency

Token usage, cache effects, and run cost estimates stay visible so teams can reason about budget and drift over time.

🔍

Change control + replayability

Review write hunks before approval, compare sessions with akmon diff, and validate deterministic behavior with replay workflows.

📈

Reliability gates

Use akmon slo verify and akmon slo trend to enforce reliability thresholds in CI instead of relying on ad hoc confidence.

🛡️

Governed rollout profiles

Move from dev to staging to prod using deterministic profile + policy-pack layering and effective-policy inspection.

// v2.0.0 session evidence model

Operate with confidence

v2.0.0 organizes Akmon around a reviewable session lifecycle: run, inspect, replay, diff, verify, and evidence export. Sessions are content-addressed, cryptographically linked, replayable, and portable as AGEF bundles.

akmon --yes --output json --task "run tests and summarize" | tee run.json
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
akmon slo trend .akmon/evidence/<session-id>.json --baseline-dir .akmon/evidence/history --window 20 --strict

Wondering how Akmon differs from other coding agents? Other tools vs Akmon — short notes in the docs, not a billboard here.

// get started

Install in seconds

GitHub Releases attach a real binary per platform. Writing to /usr/local/bin usually requires sudo on macOS — installing to ~/bin avoids that. If a download is HTML instead of a binary, the matching release asset is missing (cut a v* tag after the release workflow runs).

# Apple Silicon — M1, M2, M3, M4 (no sudo)
mkdir -p ~/bin
curl -L https://github.com/radotsvetkov/akmon/releases/latest/download/akmon-darwin-arm64 \
  -o ~/bin/akmon && chmod +x ~/bin/akmon
# If `akmon` is not found, add ~/bin to PATH (zsh):
# echo 'export PATH="$HOME/bin:$PATH"' >> ~/.zshrc && source ~/.zshrc
# Alternative: sudo curl … -o /usr/local/bin/akmon && sudo chmod +x /usr/local/bin/akmon

akmon --version # akmon 2.0.0
akmon config # run setup wizard
cd your-project
akmon --yes --task "summarize failing tests and propose minimal fixes"
mkdir -p ~/bin
curl -L https://github.com/radotsvetkov/akmon/releases/latest/download/akmon-darwin-x86_64 \
  -o ~/bin/akmon && chmod +x ~/bin/akmon

akmon --version
mkdir -p ~/bin
curl -L https://github.com/radotsvetkov/akmon/releases/latest/download/akmon-linux-x86_64 \
  -o ~/bin/akmon && chmod +x ~/bin/akmon

akmon --version
git clone https://github.com/radotsvetkov/akmon
cd akmon

# Slim build — 3.4MB, no semantic indexing
cargo build --release --no-default-features

# Full build — 20MB, with semantic indexing
cargo build --release

mkdir -p ~/bin
cp target/release/akmon ~/bin/
# 1. Install Ollama from https://ollama.com
ollama pull qwen2.5-coder:7b

# 2. Install Akmon (example: Apple Silicon → ~/bin)
mkdir -p ~/bin
curl -L https://github.com/radotsvetkov/akmon/releases/latest/download/akmon-darwin-arm64 \
  -o ~/bin/akmon && chmod +x ~/bin/akmon

# 3. Open any project
cd my-project
akmon chat # free, offline, private
// providers

Works with
every provider

One tool, any model. Set an environment variable and it works. No config files required to get started.

Ollama
Any local model. Llama, Qwen, DeepSeek, Mistral. Fully offline.
free
Anthropic
Claude Haiku, Sonnet, Opus. Direct API access with prompt caching.
BYOK
OpenRouter
500+ models. One key. Automatic provider failover. 5.5% platform fee.
BYOK
Groq
800+ tokens/second. Llama 3.3 70B. Fastest inference available.
BYOK
OpenAI
GPT-4o, GPT-4o-mini. Full tool use support.
BYOK
Azure OpenAI
Enterprise GPT-4o via Azure. Private networking, compliance support.
enterprise
Amazon Bedrock
Claude and Llama via AWS. Stays inside your VPC. SigV4 auth.
enterprise
Any OpenAI-compatible
LM Studio, Together AI, Mistral, DeepSeek, or your own endpoint.
BYOK
// trust model

You see
everything

Akmon logs every action to a JSONL audit file. Every session ends with a full summary—so you can see what ran, what it cost, and what changed.

  • 01

    Colored diff before every write

    See exactly what changes before approving. Green for additions, red for removals.

  • 02

    Per-session JSONL audit log

    Every tool call, permission verdict, and token usage logged with timestamps.

  • 03

    Exit summary with cost breakdown

    Sessions end with tool calls, files touched, cache savings, and USD cost.

  • 04

    Real-time cost in status bar

    Cache hit tokens shown in green. Session cost always visible.

session summary
Akmon session complete

Session
──────────────────────────────
ID a1b2c3d4
Duration 12m 34s
Model claude-haiku-4-5

Activity
──────────────────────────────
Messages 14
Tool calls 23
✓ Succeeded 22
✗ Failed 1
Files written 3

Tokens
──────────────────────────────
Input 48,291
Output 6,847
Cache hit 41,203 (85% savings)
Est. cost ~$0.047

.akmon/audit/a1b2c3d4.jsonl

Agent powering down. Goodbye!
// community

Built in the open

Akmon is Apache 2.0—friendly to shipping CLI wrappers, internal forks, and agent experiments. Issues, docs fixes, and focused PRs are welcome.

Contribute

Read the development setup, run tests with cargo test --workspace, and open a PR with a clear description.

Documentation

The book covers tutorials, capabilities, and automation patterns—help us keep examples sharp.

License

Apache 2.0 only. See license notes for why that fits agent tooling.

// get started

Ship with confidence.

The AI agent that respects your security, your budget, and your workflow. Single binary. No subscription. Every action audited.

↓ Install Akmon Read docs → Tutorials → Reviewer flow → GitHub →