emerik keeps the decisions behind your code honest as agents and teams multiply. Knowledge is captured as your engineers commit, contracts are enforced in your CI, and concurrent work coordinates itself — with no server, no database, and nothing outside your repository.
One engineer with one agent is productive. Fifty engineers with hundreds of agent sessions are something else entirely — unless the knowledge they run on is shared, current, and enforced.
The reasoning behind code — why a value is what it is, which invariant must hold, what was tried and rejected — is lost between sessions, between agents, and between people. Every new conversation starts from zero.
Documentation rots silently. A decision written down in January is contradicted by the code in March, and nothing notices. On measured enterprise projects, drift-driven rework consumes 5–10% of total effort.
AI multiplies the number of hands in a codebase. Without shared memory and clear territory, concurrent agents and teams overwrite each other's work and re-litigate settled decisions.
Every existing fix adds a service. emerik adds a directory.
Every change deposits the reasoning behind it — automatically, with nothing extra for anyone to write.
When the work quietly diverges from an agreed decision, emerik surfaces it — and can stop the release until it's resolved.
Fifty engineers and their agents stay coherent on one shared, self-healing source of truth.
emerik lives inside the repository as an append-only knowledge substrate — versioned by the same git history as the code it describes. Everything else follows from four concepts.
Knowledge captured automatically as your engineers commit, bound to the exact source it describes. When that source changes, the note is re-derived; when the source goes away, the note is retired. Nothing goes stale silently.
A decision written as a contract that carries a mechanical check. When reality diverges from the decision, the drift surfaces at the severity you chose — block warn advisory — up to failing the build.
A time-bounded, exclusive claim on a territory of the codebase, so concurrent engineers and agents never trample each other — coordinated entirely through git, with no lock server to run.
An append-only knowledge graph that travels with the code. A change supersedes; a deletion leaves a tombstone. The full decision history of the project stays replayable forever.
The load-bearing promise: a check that guards a decision must be able to fail when the decision is broken. A drift detector that can never detect drift only lends false confidence — emerik's checks are exercised in both directions.
Planning, engineering, review, and coordination usually mean four disconnected tools. emerik replaces the stack with one substrate the whole delivery organization — human and agent — works against.
Every commit harvests the reasoning behind the change into the substrate, automatically. There is no separate documentation step to skip, and stale notes expire on their own.
Decisions carry real, mechanical checks. When the code diverges from a decision, your CI pipeline says so — before a human has to notice, and before the drift compounds into rework.
Briefs, requirements, and architecture decisions are authored into the same substrate the code ships with — never a document that rots in a wiki. Existing codebases are seeded into the graph in an afternoon.
Territory claims, work trails, and stewardship let many engineers and their agents work in parallel on the same codebase — no coordination server, no standing meeting, no merge-day surprise.
Risk-tiered review routes routine changes through an agent quorum with four adversarial reviewer roles, and escalates only what deserves a human. Every caught failure can be minted into a durable constraint — the same mistake never ships twice.
Agents reach the substrate only through a governed tool boundary — thirty-five tools and fourteen workflow skills. Claude Code and Cursor are wired up in a single setup step, and both coordinate through the same substrate.
emerik was built by Bloomteq to run its own enterprise delivery — under the same security, sovereignty, and audit constraints your organization carries.
Knowledge lives in your git repository and never leaves your remote. Retrieval and semantic search run locally — no query and no content is sent to any third party.
No server, no database, no telemetry, no vendor endpoint. emerik runs entirely inside your infrastructure, behind your existing access controls, wherever your repositories already live.
The substrate is append-only: every change supersedes its predecessor, every deletion leaves a tombstone. The complete decision history of a project replays on demand — who decided what, when, and what replaced it.
Drift-driven rework is measurable — and gated. Organizations running emerik target roughly 2% rework against the 5–10% typical of AI-accelerated delivery at scale.