{"clean-is-not-ready-to-spend":{"description":"A valid AI workflow can still waste money. Nika now separates legal files from workflows that are ready for a paid run.","file":"2026-08-21-clean-is-not-ready-to-spend.md","author":"Thibaut Melen"},"arm64-ai-workflows-on-linux-servers":{"description":"A practical path from a native ARM64 Nika binary to checked, sandboxed and traceable AI workflows on a headless Linux host.","file":"2026-08-21-arm64-ai-workflows-on-linux-servers.md","author":"Thibaut Melen"},"the-access-layer-is-an-operating-contract":{"description":"From v0.108.0 to v0.111.0, Nika separated the model you choose from the path that reaches it, cut the workflow language to its durable core, added an explicit scheduler and turned check into an authoring loop.","file":"2026-08-20-the-access-layer-is-an-operating-contract.md","author":"Thibaut Melen"},"workflow-ecosystem-field-guide":{"description":"Choose deliberately among model providers, local runtimes, builtins, MCP servers and coding-agent surfaces without turning the workflow into vendor glue.","file":"2026-08-19-workflow-ecosystem-field-guide.md","author":"Thibaut Melen"},"coding-agents-write-plans":{"description":"Codex, Claude Code, Cursor and other coding agents can author Nika workflows while the checker, the human review and the runner keep separate authority.","file":"2026-08-17-coding-agents-write-plans.md","author":"Thibaut Melen"},"ci-without-ambient-authority":{"description":"Run Nika in CI and on headless servers with the same checked workflow bytes, explicit authority, deterministic rehearsals and a hard launch budget.","file":"2026-08-15-ci-without-ambient-authority.md","author":"Thibaut Melen"},"retire-the-prompt-glue":{"description":"A practical migration from brittle shell and Python prompt chains to a checked workflow whose graph, authority, cost and outputs are visible before it runs.","file":"2026-08-13-retire-the-prompt-glue.md","author":"Thibaut Melen"},"intent-as-code-engineering-discipline":{"description":"The useful part of AI work is not the chat. It is the intent, boundary, plan and evidence that a team can review before and after a run.","file":"2026-08-11-intent-as-code-engineering-discipline.md","author":"Thibaut Melen"},"the-run-keeps-its-receipt":{"description":"A trace is more than a log: it can replay without effects, resume verified work, expose one task, export to OpenTelemetry and assemble a redacted evidence pack.","file":"2026-08-08-the-run-keeps-its-receipt.md","author":"Thibaut Melen"},"the-two-clocks-behind-a-model-string":{"description":"A provider name can be part of the stable language while its models, prices and access paths keep changing. Nika resolves those truths on separate clocks and records the result.","file":"2026-08-03-the-two-clocks-behind-a-model-string.md","author":"Thibaut Melen"},"one-workflow-three-local-model-lanes":{"description":"Ollama, LM Studio and vLLM expose different operating experiences, but Nika keeps the workflow contract unchanged: one model string selects the local seat.","file":"2026-07-29-one-workflow-three-local-model-lanes.md","author":"Thibaut Melen"},"mcp-with-a-blast-radius":{"description":"An MCP tool is executable foreign capability. Put its server, network, environment, tool ids and schema pin under separate review, then let the trace record the call.","file":"2026-07-24-mcp-with-a-blast-radius.md","author":"Thibaut Melen"},"the-registry-reproves-everything":{"description":"A workflow registry where nothing is taken on faith: every entry is content-pinned to an exact source revision and re-proven by CI (hash, certificate, advisories). The npm-of-workflows, minus the trust-me.","file":"2026-07-14-the-registry-reproves-everything.md","author":"Thibaut Melen"},"written-by-agents":{"description":"nika init teaches the language to whatever agent you already run. Then two real agents write the same workflow: one invents a schema the audit refuses, one goes green first shot. The human review is a two-line diff.","file":"2026-07-11-written-by-agents.md","author":"Thibaut Melen"},"the-pipeline-is-a-file":{"description":"An llm pipeline is a graph of model calls, tools and processes. A graph is declared, not programmed: forty lines of YAML replace the orchestration framework. Diffable, auditable, replayable.","file":"2026-07-11-the-pipeline-is-a-file.md","author":"Thibaut Melen"},"the-mcp-server-you-didnt-build":{"description":"Everyone is learning how to build an MCP server. For your workflows, skip the tutorial: the nika binary already is one. Read-only tools let any agent audit a plan and learn the language from the thing that enforces it.","file":"2026-07-11-the-mcp-server-you-didnt-build.md","author":"Thibaut Melen"},"the-local-forecast":{"description":"explain --forecast computes duration and cost priors from your own recorded runs: stats over .nika/traces/, honest ranges at two runs, percentiles at five. Never a model call, never the network.","file":"2026-07-11-the-local-forecast.md","author":"Thibaut Melen"},"the-generative-workflow":{"description":"Strip the buzzword and a generative ai workflow is three things: inputs you name, a model step, and effects you can point to. Declare all three in one file and the mystery evaporates. Rehearsed offline, every asset with provenance.","file":"2026-07-11-the-generative-workflow.md","author":"Thibaut Melen"},"the-chain-of-custody":{"description":"A recorded run is a text file, and text files can be edited. nika trace verify recomputes the hash chain: one changed word in history breaks every line after it, and the run's printed head closes the loop.","file":"2026-07-11-the-chain-of-custody.md","author":"Thibaut Melen"},"the-agent-workflow-spectrum":{"description":"An ai agent workflow is any pipeline where a model chooses some of the steps. Every design sits between a declared graph and a free loop. Where yours lands decides which safety you can even ask for.","file":"2026-07-11-the-agent-workflow-spectrum.md","author":"Thibaut Melen"},"the-run-that-waits":{"description":"The approval gate is a task in the file, not a Slack thread: a plan fails closed on nika:prompt, and the human's answer rides the resume.","file":"2026-07-10-the-run-that-waits.md","author":"Thibaut Melen"},"the-resume-story":{"description":"kill -9 a run mid-flight, then resume it: finished work never runs twice. Durability as a file property, shown from the real journal.","file":"2026-07-10-the-resume-story.md","author":"Thibaut Melen"},"the-one-task-rerun":{"description":"Regenerate one block without re-running the world: --task scopes a fresh run to a task and its upstream, --from re-rolls what the hashes cannot see.","file":"2026-07-10-the-one-task-rerun.md","author":"Thibaut Melen"},"prompts-are-code":{"description":"Prompt versioning without a platform: the prompt lives in the workflow file, so git diffs it, a PR reviews it, git reverts it. The engine itself names an edited prompt when two runs diverge.","file":"2026-07-08-prompts-are-code.md","author":"Thibaut Melen"},"injection-goes-nowhere":{"description":"Prompt injection examples usually end with the agent taking a new action. Here the plan is authored before the model runs: the hostile note becomes data, never an action, and the boundary is checked before a token is spent.","file":"2026-07-08-injection-goes-nowhere.md","author":"Thibaut Melen"},"the-run-becomes-evidence":{"description":"Every journal line now carries a hash chain: verify names the first broken link, reproduce classifies every task, and the journal attests which engine wrote it. Trust, but verify. Then verify.","file":"2026-07-07-the-run-becomes-evidence.md","author":"Thibaut Melen"},"time-travel-for-real":{"description":"Every debugger replays the past by re-running it. Nika's past is a file, so stepping backward is free, breakpoints live in your YAML, and F5 debugs a run that already happened.","file":"2026-07-06-time-travel-for-real.md","author":"Thibaut Melen"},"the-editor-tells-the-truth":{"description":"A green badge is a claim. We audited the editor extension against the engine the way you'd audit a client against a server, and found the badge lying. Here is the contract that keeps it honest now.","file":"2026-07-06-the-editor-tells-the-truth.md","author":"Thibaut Melen"},"the-credentials-your-pipeline-breaks":{"description":"OpenAI and Google sign the images their APIs return. Almost every pipeline that touches those files silently converts the signature into evidence of tampering. That included ours, until this week.","file":"2026-07-06-the-credentials-your-pipeline-breaks.md","author":"Thibaut Melen"},"one-wire-five-servers":{"description":"The self-hosted media world quietly standardized on OpenAI's wire shapes. That accident of history is the best sovereignty news in years, and Nika's media builtins are built on it.","file":"2026-07-06-one-wire-five-servers.md","author":"Thibaut Melen"},"media-are-workflow-citizens":{"description":"Images and speech now render inside workflows under the same discipline as everything else: permit-gated saves, sha256 provenance, honest warnings, real cost on the ledger, and the sovereign path first.","file":"2026-07-06-media-are-workflow-citizens.md","author":"Thibaut Melen"},"the-trace-you-can-replay":{"description":"Every run leaves a flight recorder: a deterministic AI audit trail you can replay like a film, never re-execute by accident.","file":"2026-07-05-the-trace-you-can-replay.md","author":"Thibaut Melen"},"the-secrets-line":{"description":"Information-flow, audited before it flows: how the checker proves a secret cannot leak into a prompt, a file, or a host.","file":"2026-07-05-the-secrets-line.md","author":"Thibaut Melen"},"the-cost-line":{"description":"The audit prices every token before it is spent: ceilings on generations, budgets on loops, and a $0.00 that means it.","file":"2026-07-05-the-cost-line.md","author":"Thibaut Melen"},"anatomy-of-a-verb":{"description":"infer, exec, invoke, agent: what makes the four verbs genuinely different execution models, in the engine's own verdicts.","file":"2026-07-05-anatomy-of-a-verb.md","author":"Thibaut Melen"},"own-your-stack":{"description":"One Rust binary, your models, your files. Run LLMs locally and see what local-first actually buys you.","file":"2026-07-02-own-your-stack.md","author":"Thibaut Melen"},"intent-as-code":{"description":"Chats evaporate, files compound. The case for writing AI work down as source you own: your best prompts, versioned like code.","file":"2026-06-15-intent-as-code.md","author":"Thibaut Melen"},"dag-for-free":{"description":"The wiring is all you write: with: for data, after: for order. The plan falls out of the file: parallel waves, drawn before anything runs.","file":"2026-06-12-dag-for-free.md","author":"Thibaut Melen"},"blast-radius-in-the-file":{"description":"The workflow declares its effect boundary before execution. Static checks explain the fit; runtime sinks enforce what the host can prove.","file":"2026-06-12-blast-radius-in-the-file.md","author":"Thibaut Melen"},"standard-library-not-plugin-store":{"description":"Builtins ship with the engine and stay separate from the four verbs. External tools remain explicit dependencies, not new grammar.","file":"2026-05-27-standard-library-not-plugin-store.md","author":"Thibaut Melen"},"open-spec-copyleft-engine":{"description":"The language contract is Apache-2.0. The reference engine is AGPL-3.0-or-later. The split keeps adoption and implementation custody separate.","file":"2026-05-22-open-spec-copyleft-engine.md","author":"Thibaut Melen"},"four-verbs":{"description":"A verb is a distinct execution model, not a feature. Why the language locks at four, forever.","file":"2026-05-22-four-verbs.md","author":"Thibaut Melen"},"starting-over-on-purpose":{"description":"We had a working prototype. We rebuilt from scratch anyway: craft, not extraction, one gate at a time.","file":"2026-04-14-starting-over-on-purpose.md","author":"Thibaut Melen"},"naming-the-drum":{"description":"Why a local workflow engine carries the name of a liberation figure, and why a butterfly signs the work.","file":"2026-03-21-naming-the-drum.md","author":"Thibaut Melen"},"the-part-git-cannot-prove":{"description":"Then the public record goes quiet. An honest origin story keeps the missing months visible instead of inventing a montage.","file":"2026-02-the-part-git-cannot-prove.md","author":"Thibaut Melen"},"what-survived-the-first-version":{"description":"Strip away the early claims and four technical bets remain: file, graph, model choice and a local engine.","file":"2026-01-what-survived-the-first-version.md","author":"Thibaut Melen"},"the-first-page-was-too-sure":{"description":"On December 30, Nika became public and immediately overclaimed. The first page preserves both the idea and the correction.","file":"2025-12-30-the-first-page-was-too-sure.md","author":"Thibaut Melen"},"the-file-came-before-the-language":{"description":"The first durable Nika decision fit in two words: a file. Syntax, graphs and providers came later.","file":"2025-11-the-file-came-before-the-language.md","author":"Thibaut Melen"},"the-note-that-started-it":{"description":"Nika did not begin with a parser. It began with an objection: useful AI work should not disappear when a chat closes.","file":"2025-10-17-the-note-that-started-it.md","author":"Thibaut Melen"}}
the journal
Notes from the source.

Long-form pedagogy on Intent as Code: why useful AI work belongs in a file, why the language locks at four verbs, and what local-first actually buys you. The archive runs from the first note to today. Retrospectives keep their story date and show when they were actually published.
49 live · daily queue
cadenceDaily, human-reviewed
latest2026-08-21
methodResearch → check → publish
latest · Engine|Security · 2026-08-21Clean is not ready to spendA valid AI workflow can still waste money. Nika now separates legal files from workflows that are ready for a paid run.6 min read
this blog is a folder · content/blog on GitHub ↗ · PRs welcome · rss
Intent as Code · 4 verbs · 28 builtins · 17 providers · written down, replayable, yours