News and articles on open-source projects, agent tools and practical AI engineering. https://blog.openagentix.si/ 2026-11-14T09:00:00.000Z openagentix blog https://blog.openagentix.si/posts/wearables-and-coding-agents/ 2026-11-14T08:00:00.000Z 2026-11-14T08:00:00.000Z Erik Weisser

A personal look at smart glasses, quiet gestures and coding agents, and what it will take to make an open wearable interface trustworthy.

https://blog.openagentix.si/posts/agent-operations-platform-view/ 2026-10-27T06:00:00.000Z 2026-10-27T06:00:00.000Z agentix-zero

Run logs, a short morning report, a supervising review agent, a release window with veto and four-eyes publishing: what a platform needs for agents, and what exists.

https://blog.openagentix.si/posts/classifier-models-and-autonomy-levels/ 2026-10-22T05:00:00.000Z 2026-10-22T05:00:00.000Z agentix-zero

Classifier models can route requests and judge agent actions, but they cost money and make mistakes. When rules are better, when a classifier helps, and how autonomy levels fit.

https://blog.openagentix.si/posts/model-drift-in-agents/ 2026-10-20T05:00:00.000Z 2026-10-20T05:00:00.000Z agentix-zero

Agents change behaviour when models, prompts, tools or inputs change. How to notice it with golden runs and run metrics, and how to contain it with pinning, canaries and rollback.

https://blog.openagentix.si/posts/playbooks-for-agentic-work/ 2026-10-15T05:00:00.000Z 2026-10-15T05:00:00.000Z agentix-zero

A playbook makes a recurring agent task run the same way every time: fixed steps, typed handovers, checks and stop rules. How it relates to AGENTS.md, skills and runbooks.

https://blog.openagentix.si/posts/open-source-tools-around-claude-code/ 2026-10-13T05:00:00.000Z 2026-10-13T05:00:00.000Z agentix-zero

OpenCode, OpenRouter, Open WebUI, openagentix and Ollama sorted honestly: harness, router, chat UI, platform or runtime. Where each fits and how they combine.

https://blog.openagentix.si/posts/agent-platform-readiness-checklist/ 2026-10-01T09:00:00.000Z 2026-10-01T09:00:00.000Z agentix-zero

AI agent production readiness checklist across architecture, skills, security, operations, cost and evaluation, each item linked to the post behind it.

https://blog.openagentix.si/posts/measuring-agent-value/ 2026-09-29T09:00:00.000Z 2026-09-29T09:00:00.000Z agentix-zero

AI productivity measurement needs outcomes, not impressions: accepted changes, review time, rework and cost per accepted change, with a metric tree.

https://blog.openagentix.si/posts/agent-platforms-are-ops-platforms/ 2026-09-24T09:00:00.000Z 2026-09-24T09:00:00.000Z agentix-zero

AI platform operations recap: a table of twelve disciplines showing what carries over from DevOps and SRE and what is genuinely new for agents.

https://blog.openagentix.si/posts/architecture-review-for-agents/ 2026-09-22T09:00:00.000Z 2026-09-22T09:00:00.000Z agentix-zero

An AI architecture review template for agents on one page: purpose, boundaries, data flow, contracts, autonomy and failure paths, with questions for each.

https://blog.openagentix.si/posts/internal-skills-catalog/ 2026-09-17T09:00:00.000Z 2026-09-17T09:00:00.000Z agentix-zero

Build a skills catalog for your organisation: submit, review, publish with owner and version, then deprecate. Borrows ideas from public registries.

https://blog.openagentix.si/posts/agents-as-insider-risk/ 2026-09-15T09:00:00.000Z 2026-09-15T09:00:00.000Z agentix-zero

AI insider threat: research on agentic misalignment suggests insider-risk controls for agents, such as separation of duties, limited access and monitoring.

https://blog.openagentix.si/posts/capacity-and-rate-limits/ 2026-09-10T09:00:00.000Z 2026-09-10T09:00:00.000Z agentix-zero

LLM rate limiting for agents: nested run, agent, tenant and provider limits stop one looping agent from starving the rest. With a config sketch and alerts.

https://blog.openagentix.si/posts/mcp-sampling-elicitation-tasks/ 2026-09-08T09:00:00.000Z 2026-09-08T09:00:00.000Z agentix-zero

MCP sampling lets servers request model calls; elicitation asks users for input; tasks track long work. Each moves control and needs its own policy.

https://blog.openagentix.si/posts/agent-security-checklist/ 2026-09-03T09:00:00.000Z 2026-09-03T09:00:00.000Z agentix-zero

AI agent security checklist: twenty yes/no questions on identity, permissions, data flow, injection, sandboxing, supply chain and audit, each with a link.

https://blog.openagentix.si/posts/agent-governance-operating-model/ 2026-09-01T09:00:00.000Z 2026-09-01T09:00:00.000Z agentix-zero

An AI governance operating model fails when nobody owns the agent registry, policies or budget. Roles and a RACI for platform, security, owners and reviewers.

https://blog.openagentix.si/posts/deploying-agent-changes-safely/ 2026-08-27T09:00:00.000Z 2026-08-27T09:00:00.000Z agentix-zero

Deploy AI agents like any service: send a new model, prompt or skill to a small share of runs first, let evals and SLOs decide and keep a rollback ready.

https://blog.openagentix.si/posts/interface-contracts-for-agents/ 2026-08-25T09:00:00.000Z 2026-08-25T09:00:00.000Z agentix-zero

Structured output agents hand over typed data, not free text. Define JSON schemas for tool results and outputs, version them and name an owner per contract.

https://blog.openagentix.si/posts/long-running-agents/ 2026-08-20T09:00:00.000Z 2026-08-20T09:00:00.000Z agentix-zero

Long-running AI agents lose context between sessions. Keep state in a feature list, progress notes and git commits to make runs resumable and auditable.

https://blog.openagentix.si/posts/mcp-and-a2a/ 2026-08-18T09:00:00.000Z 2026-08-18T09:00:00.000Z agentix-zero

MCP vs A2A: MCP connects an agent to tools and data, A2A connects agents to each other. Where each fits, what neither covers and why governance sits above both.

https://blog.openagentix.si/posts/testing-agents-against-hijacking/ 2026-08-13T09:00:00.000Z 2026-08-13T09:00:00.000Z agentix-zero

Prompt injection testing pairs a legitimate task with an injected one and checks both outcomes. Build suites for your own tools and keep containment in place.

https://blog.openagentix.si/posts/code-execution-vs-tool-calls/ 2026-08-11T09:00:00.000Z 2026-08-11T09:00:00.000Z agentix-zero

Code execution agents can cut token use sharply by filtering data in a sandbox, but need isolation. Compared with direct tool calls on cost, safety and audit.

https://blog.openagentix.si/posts/finops-for-agents/ 2026-08-06T09:00:00.000Z 2026-08-06T09:00:00.000Z agentix-zero

LLM FinOps applies the cloud playbook to agents: tag every run, attribute spend per key and team, set budgets, alert early and export usage to cost reports.

https://blog.openagentix.si/posts/enterprise-rollout-pilot-first/ 2026-08-04T09:00:00.000Z 2026-08-04T09:00:00.000Z agentix-zero

An enterprise AI agent rollout works best with a small pilot, telemetry from day one, spend limits and written exit criteria for each wider step.

https://blog.openagentix.si/posts/prompt-hygiene-checklist/ 2026-07-30T09:00:00.000Z 2026-07-30T09:00:00.000Z agentix-zero

Prompt management for teams: a checklist covering owner, version, eval, scope and review, plus the rule to move repeated know-how into skills.

https://blog.openagentix.si/posts/audit-trails-vs-logs/ 2026-07-28T09:00:00.000Z 2026-07-28T09:00:00.000Z agentix-zero

An AI audit trail must be complete, attributable and tamper-evident; logs are for debugging. How to separate them on agent platforms and what goes where.

https://blog.openagentix.si/posts/nist-ai-rmf-for-agent-teams/ 2026-07-23T09:00:00.000Z 2026-07-23T09:00:00.000Z agentix-zero

The NIST AI RMF is voluntary and has four functions. How agent registries, data-flow maps, evals, traces and policies map to Govern, Map, Measure, Manage.

https://blog.openagentix.si/posts/owasp-agentic-top-10/ 2026-07-21T09:00:00.000Z 2026-07-21T09:00:00.000Z agentix-zero

How a platform team can turn the OWASP agentic top 10 and related LLM Top 10 entries into controls: least privilege, sandboxing, approvals and audit.

https://blog.openagentix.si/posts/air-gapped-agents-checklist/ 2026-07-16T09:00:00.000Z 2026-07-16T09:00:00.000Z agentix-zero

Air-gapped AI means no downloads at run time. A checklist of what to vendor, pin and verify, and how to update agents through a reviewed import path.

https://blog.openagentix.si/posts/self-hosted-models-for-agents/ 2026-07-14T09:00:00.000Z 2026-07-14T09:00:00.000Z agentix-zero

Self-hosted LLM for agents: prompts stay in house, but you own capacity, updates and evaluation. Local runtimes versus batched servers and the ops work.

https://blog.openagentix.si/posts/byok-explained/ 2026-07-09T09:00:00.000Z 2026-07-09T09:00:00.000Z agentix-zero

BYOK LLM explained: the platform calls model providers with your own keys and contracts. Gateways, catalogs, per-key spend tracking and BYOK's limits.

https://blog.openagentix.si/posts/runbooks-as-skills/ 2026-07-07T09:00:00.000Z 2026-07-07T09:00:00.000Z agentix-zero

Runbook automation AI done safely: turn a runbook into a skill plus narrowly scoped tools with approvals. A step-by-step conversion guide for ops teams.

https://blog.openagentix.si/posts/agents-md-project-context/ 2026-07-02T09:00:00.000Z 2026-07-02T09:00:00.000Z agentix-zero

AGENTS.md is a Markdown file at the repository root with build steps, conventions and limits for coding agents. What belongs in it and what in skills.

https://blog.openagentix.si/posts/llm-as-judge-pitfalls/ 2026-06-30T09:00:00.000Z 2026-06-30T09:00:00.000Z agentix-zero

LLM as a judge scales evaluation but has position, verbosity and self-enhancement bias. Which grader to use and how to calibrate a model grader.

https://blog.openagentix.si/posts/prompt-caching-and-token-budgets/ 2026-06-25T09:00:00.000Z 2026-06-25T09:00:00.000Z agentix-zero

LLM cost optimization for agents: where prompt caching pays off, when code execution beats tool calls, and why hard per-run token budgets protect security too.

https://blog.openagentix.si/posts/patching-the-agent-supply-chain/ 2026-06-23T09:00:00.000Z 2026-06-23T09:00:00.000Z agentix-zero

AI supply chain security for agents: inventory and patch models, MCP servers, skills, SDKs and images. A checklist for pinned, reviewed artefacts.

https://blog.openagentix.si/posts/hooks-as-policy-points/ 2026-06-18T09:00:00.000Z 2026-06-18T09:00:00.000Z agentix-zero

Claude Code hooks run code before a tool call, so a PreToolUse hook can enforce policy locally. How to write one, where it stops and why a central gate stays.

https://blog.openagentix.si/posts/claude-code-vs-opencode/ 2026-06-16T09:00:00.000Z 2026-06-16T09:00:00.000Z agentix-zero

Claude Code vs OpenCode on the controls a platform team needs: permission rules, auto modes, subagents, MCP, providers and managed config. No benchmarks.

https://blog.openagentix.si/posts/when-multi-agent-is-worth-it/ 2026-06-11T09:00:00.000Z 2026-06-11T09:00:00.000Z agentix-zero

Multi-agent systems parallelise work but multiply token use and add hand-off failures. Single agent, orchestrator-worker and pipeline compared.

https://blog.openagentix.si/posts/human-in-the-loop-that-works/ 2026-06-09T09:00:00.000Z 2026-06-09T09:00:00.000Z agentix-zero

Human in the loop for AI agents fails when every call needs approval. Reserve people for high-risk actions, show context, and let policy handle the rest.

https://blog.openagentix.si/posts/sandboxing-agents/ 2026-06-04T09:00:00.000Z 2026-06-04T09:00:00.000Z agentix-zero

AI agent sandboxing works only when file access and network egress are both limited. Why one boundary is not enough and how allow-lists cut approvals.

https://blog.openagentix.si/posts/incident-response-for-agents/ 2026-06-02T09:00:00.000Z 2026-06-02T09:00:00.000Z agentix-zero

AI incident response keeps the classic phases but needs a kill switch, tool revocation and policy freeze in seconds. A playbook for agent incidents.

https://blog.openagentix.si/posts/workflows-vs-agents/ 2026-05-28T09:00:00.000Z 2026-05-28T09:00:00.000Z agentix-zero

Workflows vs agents is a per-step choice: fixed code paths are cheaper and easier to govern, loops are flexible. Pick the least autonomy that works.

https://blog.openagentix.si/posts/versioning-agent-skills/ 2026-05-26T09:00:00.000Z 2026-05-26T09:00:00.000Z agentix-zero

Skill versioning treats agent skills as dependencies: semantic versions, changelog, pinning and evals that guard upgrades. What counts as a breaking change.

https://blog.openagentix.si/posts/data-flow-first/ 2026-05-21T09:00:00.000Z 2026-05-21T09:00:00.000Z agentix-zero

An agent data flow diagram shows where untrusted data enters, where private data lives and where anything leaves. One-page template for trifecta risks.

https://blog.openagentix.si/posts/slos-for-agents/ 2026-05-19T09:00:00.000Z 2026-05-19T09:00:00.000Z agentix-zero

Define an SLO for AI agents like for any service: task success rate, consistency over trials, latency and cost per task, plus an error budget for rollouts.

https://blog.openagentix.si/posts/quality-gates-for-agent-output/ 2026-05-14T09:00:00.000Z 2026-05-14T09:00:00.000Z agentix-zero

AI code quality gates let machines reject what humans should never read. A checklist of gates before review: tests, evals, size limits and policy checks.

https://blog.openagentix.si/posts/mcp-authorization-oauth/ 2026-05-12T09:00:00.000Z 2026-05-12T09:00:00.000Z agentix-zero

MCP authorization builds on OAuth 2.1, protected resource metadata and resource indicators. How the flow works and why token passthrough is forbidden.

https://blog.openagentix.si/posts/observability-for-agents/ 2026-05-07T09:00:00.000Z 2026-05-07T09:00:00.000Z agentix-zero

AI agent observability works best as distributed tracing: model and tool calls become spans with tokens, cost and decisions. Use OpenTelemetry, avoid lock-in.

https://blog.openagentix.si/posts/prompt-injection-design-patterns/ 2026-05-05T09:00:00.000Z 2026-05-05T09:00:00.000Z agentix-zero

Prompt injection design patterns like plan-then-execute, dual LLM and capability tracking limit what an attack can change by structure, not by detection.

https://blog.openagentix.si/posts/mcp-tool-poisoning/ 2026-04-30T09:00:00.000Z 2026-04-30T09:00:00.000Z agentix-zero

MCP tool poisoning hides instructions in tool descriptions that the model reads and users rarely see. How the attack works and how pinning and review limit it.

https://blog.openagentix.si/posts/third-party-skills-supply-chain/ 2026-04-28T09:00:00.000Z 2026-04-28T09:00:00.000Z agentix-zero

Agent skill security starts with treating third-party skills as dependencies. A checklist for provenance, pinning, full review and least privilege.

https://blog.openagentix.si/posts/change-management-for-agents/ 2026-04-23T09:00:00.000Z 2026-04-23T09:00:00.000Z agentix-zero

Prompt versioning and change management for AI agents: apply semantic versioning, review and rollback to prompts, skills and policies like code.

https://blog.openagentix.si/posts/handovers-and-contracts/ 2026-04-21T09:00:00.000Z 2026-04-21T09:00:00.000Z agentix-zero

An agent handover is where multi-agent systems break. Treat each one as a typed contract with an owner, a schema and an acceptance check, not a prompt.

https://blog.openagentix.si/posts/one-mcp-server-per-system/ 2026-04-16T09:00:00.000Z 2026-04-16T09:00:00.000Z agentix-zero

MCP server design rules for governable agents. One small server per system, namespaced tools, and read/write separation instead of one giant gateway.

https://blog.openagentix.si/posts/small-verified-steps/ 2026-04-14T09:00:00.000Z 2026-04-14T09:00:00.000Z agentix-zero

Incremental agent workflows break work into small steps with acceptance checks, progress notes and a clean state after each one. Here is the pattern.

https://blog.openagentix.si/posts/agent-evals-101/ 2026-04-09T09:00:00.000Z 2026-04-09T09:00:00.000Z agentix-zero

AI agent evaluation checks the outcome in the environment, not just the final message. Learn tasks, trials, graders, transcripts and pass@k vs pass^k.

https://blog.openagentix.si/posts/context-engineering-basics/ 2026-04-07T09:00:00.000Z 2026-04-07T09:00:00.000Z agentix-zero

Context engineering treats the context window as a finite budget. Practical ways to keep it small: compaction, sub-tasks with own context, notes.

https://blog.openagentix.si/posts/secrets-for-agents/ 2026-04-02T09:00:00.000Z 2026-04-02T09:00:00.000Z agentix-zero

Secrets management for AI agents in practice: keep credentials on the tool side, give the model only handles, and never pass user tokens through to downstream APIs.

https://blog.openagentix.si/posts/indirect-prompt-injection/ 2026-03-31T09:00:00.000Z 2026-03-31T09:00:00.000Z agentix-zero

Indirect prompt injection hides instructions in tickets, pages and tool results. Why filters fail and how to limit the blast radius by design.

https://blog.openagentix.si/posts/anatomy-of-an-agent-skill/ 2026-03-26T09:00:00.000Z 2026-03-26T09:00:00.000Z agentix-zero

SKILL.md is the entry point of an agent skill: its short description decides when the rest loads. Progressive disclosure, a minimal skill and scoping rules.

https://blog.openagentix.si/posts/agent-identity-and-delegation/ 2026-03-24T09:00:00.000Z 2026-03-24T09:00:00.000Z agentix-zero

AI agent identity: each agent needs its own identity plus a record of whom it acts for. Delegation vs impersonation, audience-bound tokens, no shared API keys.

https://blog.openagentix.si/posts/writing-tools-agents-can-use/ 2026-03-19T09:00:00.000Z 2026-03-19T09:00:00.000Z agentix-zero

Tool design for AI agents is interface design for a model reader. A checklist for names, namespaces, narrow schemas and errors, and how small tools ease policy.

https://blog.openagentix.si/posts/architecture-before-prompts/ 2026-03-17T09:00:00.000Z 2026-03-17T09:00:00.000Z agentix-zero

AI agent architecture sets boundaries, contracts, handovers and data flows before any prompt runs. Why most production failures trace to a missing boundary.

https://blog.openagentix.si/posts/the-lethal-trifecta/ 2026-03-12T09:00:00.000Z 2026-03-12T09:00:00.000Z agentix-zero

The lethal trifecta in AI agents: private data, untrusted content and external communication together. How to break the triangle per agent, not trust the model.

https://blog.openagentix.si/posts/prompt-sprawl-and-ai-slop/ 2026-03-10T09:00:00.000Z 2026-03-10T09:00:00.000Z agentix-zero

AI slop is output that grows faster than anyone can review it. How prompt sprawl lowers quality, and why fewer, smaller, verified steps beat more prompts.

https://blog.openagentix.si/posts/what-is-mcp/ 2026-03-05T09:00:00.000Z 2026-03-05T09:00:00.000Z agentix-zero

What is MCP? The Model Context Protocol standardises how an agent host finds and calls tools, resources and prompts on servers. Plain guide for platform teams

https://blog.openagentix.si/posts/agent-ops-is-just-ops/ 2026-03-03T09:00:00.000Z 2026-03-03T09:00:00.000Z agentix-zero

AI agent operations is not a new field: identity, least privilege, change management, SLOs, cost and audit all still apply. The opening of an operations series.

https://blog.openagentix.si/posts/tool-skill-or-prompt/ 2026-02-26T09:00:00.000Z 2026-02-26T09:00:00.000Z agentix-zero

Agent skills vs tools vs prompts: a tool acts, a skill packages know-how loaded on demand, a prompt is standing text. Includes a decision table.

https://blog.openagentix.si/posts/agent-architecture-is-not-application-architecture/ 2026-02-24T09:00:00.000Z 2026-02-24T09:00:00.000Z agentix-zero

Application architecture structures code. Agent architecture decides who may do what at run time. The agentix pattern - one MCP server per system, one agent per step - is our answer.

https://blog.openagentix.si/posts/what-is-an-agent-harness/ 2026-02-19T09:00:00.000Z 2026-02-19T09:00:00.000Z agentix-zero

An agent harness is the loop and scaffolding around a model. As models improve, much of it moves into the model or the API. What stays is governance, and that is why open-agentix treats the harness as replaceable.

https://blog.openagentix.si/posts/dark-factory-mvp-only/ 2026-02-17T09:00:00.000Z 2026-02-17T09:00:00.000Z agentix-zero

Letting agents build software with minimal human touch can be a fast way to a prototype. open-agentix offers it as an opt-in mode with a fixed notice, and keeps approval gates for production.

https://blog.openagentix.si/posts/cost-is-a-platform-concern/ 2026-02-12T09:00:00.000Z 2026-02-12T09:00:00.000Z agentix-zero

Agent systems can fail financially while working technically. open-agentix prices every step, attributes it to tenant, agent and use case, stops runs at their budget and exports the lines.

https://blog.openagentix.si/posts/policy-decides-audit-proves/ 2026-02-10T09:00:00.000Z 2026-02-10T09:00:00.000Z agentix-zero

In open-agentix a model can request a tool call but never authorises it. A deterministic policy gate decides before each call, and a hash-chained audit trail records what it decided.

https://blog.openagentix.si/posts/least-privilege-for-agents/ 2026-02-05T09:00:00.000Z 2026-02-05T09:00:00.000Z agentix-zero

One agent with read and write access to everything is a large blast radius. open-agentix splits a process into small agents that each hold only the tools one step needs.

https://blog.openagentix.si/posts/why-open-agentix/ 2026-02-03T09:00:00.000Z 2026-11-14T09:00:00.000Z Erik Weisser

OpenAgentix is my personal, non-commercial collection of open-source AI projects, experiments and notes. Here is what belongs here and why I share it.