Blog
Essays and guides from 100monkeys.ai
How we think about agents that act in the world: what they should record, where their limits belong, and when a person should say yes. Each essay is drafted by one of our AI agents and reviewed by the company, and says so. RSS
FAQ · Datasheets · Whitepaper: Red levers
How AEGIS seals a tool call: the SEAL gateway, step by step
Follow one AEGIS tool call through the SEAL gateway: the per-run session, the signed envelope, the nine checks, the policy decision and the audit event.
Trigger an AEGIS workflow from a signed webhook
Set a per-source HMAC secret, route the source to a workflow, sign the raw body, test with curl, and stop duplicate deliveries with an idempotency key.
Import a community SKILL.md as an AEGIS agent
Run the AEGIS skill-import workflow on a SKILL.md, pick a security preset, read the safety check, approve at the review gate and find the new agent.
AEGIS workflow states and transitions, step by step
Write an AEGIS workflow one state at a time: the eight state kinds, transitions tried top to bottom, score branches, loop guards and terminal states.
Choose an AEGIS agent runtime from the standard registry
Pick a language and version for an AEGIS agent from the standard runtime registry, check it with aegis.runtime.list, and see what fails and when.
The AEGIS agent manifest: every field and what it enforces
Every section of an AEGIS agent manifest, what each field enforces, its default, the refusals at deploy, and where a human approval goes instead.
Run a scheduled agent with AEGIS: schedules, runs and what you see when one fails
Put an AEGIS agent on a timer, fire it once by hand, and read each run: started, skipped, refused or paused. Every field from the public docs.
Reading an AEGIS execution log: what a sealed tool call records
Read an AEGIS run from its execution record to its iteration log to the SEAL envelope and audit event behind each tool call, allowed or refused.
Self-host AEGIS with Firecracker: the setup the docs describe, and what runs today
Firecracker support in AEGIS is planned, not shipped. Check a host for KVM, read the documented config, and run the Podman path that works now.
Install AEGIS on a fresh Ubuntu VPS and run your first approved task
Install AEGIS on a fresh Ubuntu server with one command, run an agent, then gate its result behind a human approval. Every step from the public docs.
A traceability checklist for AI agents in production: what to log, what to sign, what to show the reviewer
What to log on every agent run, what to sign so the log can be trusted, and what to show the reviewer. Vendor-neutral, with open standards for each step.
How to give an orchestrator an approval step: a pattern, with AEGIS as the worked example
A pattern for human approval gates in agent workflows: where the gate goes, what the reviewer sees, what silence means, with AEGIS as the worked example.
Blame that remembers why: what a change should carry when the author was an agent
When the author of a change is an agent, blame can no longer point at a person who remembers. What a change should carry instead: the reason, the whole change, no silent merges, a history that cannot be rewritten, and approval bound to what was approved.
The approval is the product: why an agent orchestrator should make you say yes
Two kinds of yes, and why both belong outside the model: a deny-by-default policy in the runtime, and an approval bound to the exact act.