"status": "ok", "debug": { "latency_ms": 14, "trace": "gw-84f19", "cache": "hit" }, { "id": 0, "name": "item-0", "path": "src/module_0/file_0.rs" }, { "id": 1, "name": "item-1", … }, … 98 more result records … "request_id": "req-8f3a91" }Agentic OS 1.0
An operating system layer built for agents.
Cut 30–70% of your agent’s tool-output tokens with one command.
Just one thing the OS does for your agent — it also runs, recovers, and secures it.
The users of operating systems have changed. ANOLISA makes agents first-class participants at the system layer.
One entry point. Enable what you need.
curl -fsSL https://get.agentic-os.sh | bashBring your Agent in
Read https://agentic-os.sh/agents/ to learn how to use ANOLISA, then help me install it for this environment.TOKENLESS · TOOL OUTPUT COMPRESSION
Keep what matters. Send fewer tokens.
Tokenless compresses tool output before it reaches the model. Errors, changes, and useful fields stay in context. Repeated records and progress noise take up less space.
These ranges come from repository benchmarks. Results vary by payload and cannot be translated directly into provider billing savings.Read the benchmark methodology ↗
Response compression26–78%
status: ok 0,item-0,src/module_0/file_0.rs 1,item-1,src/module_1/file_1.rs … 30 more retained rows …request_id: req-8f3a91Solve the critical problems in Agent operations.
Everyday Agent operations depend on terminal collaboration, Token efficiency, and execution environments. Each capability can be installed independently and integrated on demand.
Cosh-ng
Run and take control of Agents in the terminal.
Cosh-ng runs in the Shell you already use. Give it tasks in natural language, call tools and Skills, and connect existing workflows through Hooks or MCP.
The terminal remains your workspace, and you can interrupt or resume the Agent at any time.
Start with Cosh-ng →Token Flow
See where Tokens go and reduce tool-output cost.
Tokenless compresses JSON, logs, and diffs before tool output reaches the model. AgentSight records trajectories and tracks Token usage on Linux so you can see where the context is going.
Review the compression result and retrieve original content when needed.
Start with Tokenless →Agent Runtime
Run Agent tasks with isolation and workspace recovery.
Agent Sec Core classifies commands and applies sandbox policies. ws-ckpt creates fast workspace checkpoints and rolls back file changes. SkillFS exposes selected Skills through a mounted filesystem when they are needed.
These Linux components work independently and can be combined for tasks that need stronger controls.
Start with ws-ckpt →Documentation and project resources
Find user guides, developer documentation, release history, and the machine-readable Agent entry point.