drixl 0.3.0

DRIXL - Compressed Inter-Agent Communication Language for Multi-Agent AI Systems

DRIXL
Compressed Inter-Agent Communication Language — Built for Speed, Cost, and Scale

Key Concepts · Verb Vocabulary · Message Format · Context Store · Getting Started · CLI Tool · Benchmarks

DRIXL is a compressed inter-agent communication language designed to minimize token usage and maximize value when running multiple AI agents together. Instead of agents exchanging verbose natural language, DRIXL provides a structured, minimal signal format — cutting communication overhead by up to 80%.

One standard. All agents. Zero waste.

from drixl import Message, ContextStore

# Build a compressed inter-agent message
msg = Message.build(
    to="AGT2",
    fr="AGT1",
    msg_type="REQ",
    priority="HIGH",
    actions=["ANLY", "XTRCT"],
    params=["firewall.json", "denied_ips", "out:json"],
    ctx_ref="ref#1"
)
print(msg)
# @to:AGT2 @fr:AGT1 @t:REQ @p:HIGH
# ANLY XTRCT [firewall.json] [denied_ips] [out:json] [ctx:ref#1]

Or parse an incoming DRIXL message:

from drixl import Message

raw = "@to:AGT3 @fr:AGT2 @t:RES @p:MED\nVALD [suspicious_ips:14] [src:threat_db] [out:json] [ctx:ref#1]"
parsed = Message.parse(raw)
print(parsed)
# {'envelope': {'to': 'AGT3', 'fr': 'AGT2', 'type': 'RES', 'priority': 'MED'},
#  'actions': ['VALD'], 'params': ['suspicious_ips:14', 'src:threat_db', 'out:json', 'ctx:ref#1']}

---

Key Concepts

Why DRIXL?

When multiple AI agents communicate using natural language, tokens are wasted on:

• Politeness phrases and filler words
• Redundant context repetition every message
• Verbose JSON field names
• Re-explaining roles each turn

DRIXL solves this with three layers:

• 📦 Compressed Envelope — Minimal header with routing, type, and priority
• 🗂️ Shared Context Store — Store context once, reference by ID forever
• 🔤 Verb Shortcodes — Fixed vocabulary of 4–6 letter action codes

Token Savings

Scenario	Saving
Per-message compression	~70% reduction
Shared context (no repetition)	~60% reduction
Verb vocabulary vs. prose	~40% reduction
Combined at scale	Up to 80% total

---

Message Format

Every DRIXL message has two parts — an envelope and a body:

@to:AGT2 @fr:AGT1 @t:REQ @p:HIGH
ANLY XTRCT [input_file] [output:json] [ctx:ref#3]

Envelope Fields

Field	Values	Description
@to	Agent ID	Recipient agent
@fr	Agent ID	Sender agent
@t	REQ / RES / ERR / FIN	Message type
@p	HIGH / MED / LOW	Priority

Message Types

Type	Meaning
REQ	Request — asking another agent to perform a task
RES	Response — returning results to sender
ERR	Error — reporting a failure with details
FIN	Finalize — signaling pipeline completion

---

Verb Vocabulary

DRIXL uses a fixed set of action verbs. All agents share this vocabulary:

Verb	Full Meaning	Example
ANLY	Analyze	ANLY [logs.json]
XTRCT	Extract	XTRCT [denied_ips]
SUMM	Summarize	SUMM [report.txt] [out:json]
EXEC	Execute action	EXEC [throttle_ip] [192.168.1.45]
VALD	Validate output	VALD [result] [schema:strict]
ESCL	Escalate to human	ESCL [reason:low_confidence]
ROUT	Route to agent	ROUT [AGT3] [payload:ref#5]
STOR	Save to memory	STOR [key:last_result] [val:ref#4]
FETCH	Retrieve data	FETCH [url:https://...] [out:html]
CMPX	Compare values	CMPX [val_a] [val_b] [out:diff]
RETRY	Retry task	RETRY [max:3] [delay:5s]
MERGE	Merge results	MERGE [source_a] [source_b]
SPLIT	Split for parallel	SPLIT [batch:10]
AUTH	Authenticate	AUTH [token:xyz]
LOG	Emit log entry	LOG [level:info]
WAIT	Pause execution	WAIT [condition:ready]
CACHE	Cache result	CACHE [ttl:300]

---

Context Store

Instead of repeating context in every message, DRIXL uses a shared Context Store — agents reference context by ID:

from drixl import ContextStore

store = ContextStore()  # Uses in-memory store by default (Redis supported)

# Store context once
store.set("ref#1", "Project: Network security monitoring pipeline")
store.set("ref#2", "Output format: {ip, action, timestamp, confidence}")
store.set("ref#3", "Constraints: no action if confidence < 0.85")

# Agents reference it — never repeat it
print(store.get("ref#1"))  # 'Project: Network security monitoring pipeline'

With Redis backend for multi-agent shared state:

from drixl import ContextStore

store = ContextStore(backend="redis", host="localhost", port=6379)
store.set("ref#1", "Project goal: MikroTik bandwidth monitor")

---

Getting Started

Installation

DRIXL requires Python 3.10 or higher:

pip install drixl

With CLI tools:

pip install "drixl[cli]"

With Redis context store support:

pip install "drixl[redis]"

Install everything:

pip install "drixl[all]"

Quick Example — 3 Agent Pipeline

from drixl import Message, ContextStore

# Shared context — defined once
store = ContextStore()
store.set("ref#1", "Project: Firewall threat detection")
store.set("ref#2", "Output: JSON array [{ip, count, risk_level}]")

# Agent 1 → Agent 2: Analyze logs
msg_1 = Message.build(
    to="AGT2", fr="AGT1", msg_type="REQ", priority="HIGH",
    actions=["ANLY"], params=["firewall.log", "out:json"],
    ctx_ref="ref#1"
)

# Agent 2 → Agent 3: Validate findings
msg_2 = Message.build(
    to="AGT3", fr="AGT2", msg_type="RES", priority="HIGH",
    actions=["VALD", "ROUT"], params=["findings:14_ips", "AGT3"],
    ctx_ref="ref#2"
)

# Agent 3 → Orchestrator: Done
msg_3 = Message.build(
    to="ORCH", fr="AGT3", msg_type="FIN", priority="MED",
    actions=["STOR"], params=["key:threat_report", "val:ref#7"]
)

print(msg_1)
print(msg_2)
print(msg_3)

---

CLI Tool

New in v0.2.0: DRIXL now includes a powerful command-line tool:

Parse and Validate Messages

drixl parse "@to:AGT2 @fr:AGT1 @t:REQ @p:HIGH\nANLY [input.json]"
# ✓ Valid DRIXL message
# Envelope:
#   To:       AGT2
#   From:     AGT1
#   Type:     REQ
#   Priority: HIGH
# ...

Build Messages Interactively

drixl build --to AGT2 --from AGT1 --type REQ --priority HIGH --actions ANLY,XTRCT --params input.json,out:json
# ✓ Message built successfully:
# @to:AGT2 @fr:AGT1 @t:REQ @p:HIGH
# ANLY XTRCT [input.json] [out:json]

List All Verbs

drixl verbs
# DRIXL Standard Verbs (21 total):
#   ANLY     Analyze input data or content
#   XTRCT    Extract specific fields or values
#   ...

Search Verbs

drixl verbs --search analyze
# Verbs matching 'analyze':
#   ANLY     Analyze input data or content

Benchmark Token Usage

drixl benchmark
# Token Usage Comparison
# ==================================================
# Format               Tokens    vs DRIXL    Savings
# ------------------------------------------------------
# DRIXL                    25        1.00x          -
# JSON                     60        2.40x        58%
# Natural Language        120        4.80x        79%

---

Benchmarks

Token count comparison: DRIXL vs Natural Language vs JSON vs XML for the same instruction:

#	Method	Tokens	vs DRIXL
1	DRIXL message	~25	1.0x
2	Structured JSON	~60	~2.4x
3	Natural language	~120	~4.8x
4	XML (verbose)	~140	~5.6x

Benchmarks measured using OpenAI tiktoken on 100+ real agent message samples. Use drixl benchmark to compare your own messages.

---

Project Structure

DRIXL/
├── drixl/
│   ├── __init__.py          # Public API
│   ├── message.py           # Message builder & parser
│   ├── context_store.py     # Shared context store (memory + Redis)
│   ├── verbs.py             # Standard verb vocabulary
│   ├── exceptions.py        # Custom exceptions
│   └── cli.py               # Command-line interface
├── examples/
│   ├── basic_pipeline.py    # 3-agent pipeline example
│   ├── network_monitor.py   # Network agent example
│   └── web_pipeline.py      # Web scraping agent example
├── tests/
│   ├── test_message.py
│   ├── test_context_store.py
│   ├── test_verbs.py
│   ├── test_message_enhancements.py
│   ├── test_context_ttl.py
│   └── test_cli.py
├── .github/
│   └── workflows/
│       ├── tests.yml
│       └── publish.yml
├── benchmarks.py
├── CONTRIBUTING.md
├── ROADMAP.md
├── LICENSE
├── pyproject.toml
└── README.md

---

Supporting DRIXL

If DRIXL saves you tokens and costs, consider supporting its development:

• ⭐ Star the repo — helps others discover DRIXL
• 💖 Sponsor on GitHub — fund ongoing development
• 🐛 Open an issue — report bugs or propose new verbs

---

Contributing

We welcome contributions! Please read our contributing guidelines before getting started.

[!NOTE] DRIXL is in active early development. The verb vocabulary and message format are open for community input — open an issue to propose new verbs or format extensions.

[!CAUTION] DRIXL is a communication protocol standard. Implementations using DRIXL are responsible for validating inputs and outputs. Never pass unvalidated agent outputs directly to execution functions.

License

This project is licensed under the MIT License — see LICENSE for details.

---

Designed & crafted with ❤️ by Ossama Hashim — SamoTech.