dlogs
Start free beta

DEC · append-only

Decisions your coding agents can query.

dlogs gives coding agents the decisions your team already made — so they stop rewriting what you already decided against. Mine PR history, then search with rationale and lineage over MCP.

Start free beta

Free during beta · Append-only records · Works with Cursor & Claude Code via MCP

DEC-2026-0042ACTIVE

24h token rotation window

24h window balances token freshness against Redis connection pool limits. 1h window was rejected for latency spikes under load.

Constraint
Redis pool max connections
Supersedes
DEC-2025-0018

The problem

Agents rewrite what you already decided against

Coding agents move fast. Without queryable decision context, they re-litigate settled constraints — and humans only notice in review.

  1. 01

    Agents change code without constraints

    Cursor, Claude Code, and Copilot confidently edit paths your team already constrained — unless those decisions are queryable where agents work.

  2. 02

    The decisions exist — buried in PR history

    Rationale lives in merged PRs, review threads, and people's heads. New engineers and agents cannot find it when they need it.

  3. 03

    dlogs stores key decisions, not everything

    Consequential records only: append-only DEC IDs with supersede lineage, searchable by humans and agents over MCP.

How it works

Three steps to agent context

  1. 01

    Mine your PR history

    Connect GitHub. dlogs turns merged PRs into decision drafts a human confirms — your own history on day one.

  2. 02

    Agents query over MCP

    In Cursor or Claude Code, ask what constraints govern a file. Semantic search returns the decision with rationale.

  3. 03

    Advisory context on PRs

    Agent-authored PRs get non-blocking comments citing the governing DEC — no merge gates.

Cursor · MCP

What constraints govern this file?

Agents search decisions, get decisions for a file, and receive the real DEC with rationale — before they rewrite what you already decided against.

auth/token_rotation.pyCursor · MCP
41# Token rotation strategy
42def rotate_token(user, window_hours=24):
43 """Rotate auth token within the given window."""
44 pool = get_redis_pool()
45 if pool.active_connections > pool.max - 2:
46 raise PoolExhaustedError()
47 return generate_token(user)
dlogs MCP
What constraints govern auth/token_rotation.py?
DEC-2026-0042ACTIVE

24h window balances token freshness against Redis connection pool limits. 1h window (DEC-2025-0018) was rejected for causing latency spikes under load.

Constraint: Redis pool max connections

Supersedes: DEC-2025-0018

GitHub · advisory

Advisory comments cite the governing decision

Agent-authored PRs get told which recorded decisions govern the code they touch. Checks are advisory — not merge gates.

01

The agent PR

An agent-authored PR touches a governed path.

  • Title: Widen token rotation to 1h
  • File touched: auth/token_rotation.py
  • Author: cursor-agent
02

The advisory comment

dlogs cites the decision that already governs this code.

  • Advisory check: dlogs/decision-context
  • Governing: DEC-2026-0042 (ACTIVE)
  • Constraint: 24h window — 1h was rejected for pool exhaustion
03

The human review

Reviewer sees why the change conflicts with recorded intent.

  • DEC-2026-0042 balances freshness vs Redis pool limits.
  • Supersedes: DEC-2025-0018 (1h window)
  • Non-blocking — review continues with durable context.
04

The record holds

History stays intact whether the PR updates or supersedes.

  • Option A: Align the PR to DEC-2026-0042.
  • Option B: Supersede with a new DEC and rationale.
  • Status: Append-only; no silent rewrite of history.

PR-history mining

We found the decisions your team already made

Connect GitHub. dlogs mines merged PRs into decision drafts a human confirms — so you start with your own history on day one, not an empty ADR folder.

01

Connect GitHub

Install the GitHub App on the repos that hold your history.

  • ›Select org and repositories
  • ›dlogs requests merged-PR read access
  • ›No merge gates installed
02

Mine merged PRs

dlogs extracts candidate decisions from PR titles, bodies, and review threads.

  • ›Scanning merged PRs on main
  • ›Drafts grouped by path and theme
  • ›Ready for human confirmation
03

Confirm drafts

A human reviews drafts, edits rationale, and activates records.

  • ›Draft → DEC-2026-0042
  • ›Problem, options, constraints filled from PR context
  • ›You confirm authorship and status
04

Day-one memory

Agents and humans can search decisions your team already made.

  • ›Semantic + keyword search live
  • ›MCP clients can query governing DECs
  • ›Lineage ready as you supersede over time

Lineage

Supersede preserves history

Append-only, per-org tamper-evident records. Agents and humans see the current decision in context of what came before.

An agent proposing a 1h window hits DEC-2026-0042 and sees that DEC-2025-0018 already tried it — and why it was superseded.

DEC-2025-0018superseded

1h token rotation window

@priya · Sep 2025

DEC-2026-0042active

24h token rotation window

@alex · Jan 2026

Constraint: Must not exceed Redis connection pool limit

Surfaces

Where decisions are captured, where agents find them

Capture

  • PR-history mining

    Connect GitHub; dlogs mines merged PRs into drafts you confirm.

  • Web app

    Author structured decision records with clear authorship and rationale.

  • MCP (propose)

    Agents can propose drafts with provenance; humans approve by default.

Retrieve

  • MCP

    Primary for agents: search decisions, get decisions for a file, retrieve lineage.

  • Web app

    Search and browse decisions, rationale, status, and supersede history.

  • GitHub (advisory)

    Non-blocking checks and comments cite governing decisions on PRs.

  • Slack & Linear

    Supporting capture surfaces — not the launch narrative.

Trust

Append-only records you can trust

01

Append-only by design

Decisions are never edited or deleted in place. New facts supersede old ones; history stays intact.

02

Per-org tamper-evident records

Each org gets DEC-YYYY-NNNN records with integrity controls. That is the ceiling of trust language we claim.

03

AI proposes, humans author

Agents may propose drafts with provenance. Humans approve and own accountability by default.

Direction

Agent context first. Trusted history underneath.

dlogs focuses on the useful foundation: mine the decisions you already made, keep them append-only, and put them where coding agents look.

  • Ask agents what constraints govern a file — and get a real DEC with rationale.
  • Start with mined PR history instead of an empty ADR folder.
  • Keep Slack and Linear as supporting capture — not the product identity.

FAQ

Questions

Is dlogs free right now?

Yes. dlogs is in free beta for everyone who wants to use and experiment with it.

Is there a pricing plan right now?

No. We are not charging during beta. We want teams to use dlogs, shape workflows, and help us learn before packaging decisions.

What is dlogs in one line?

dlogs gives coding agents the decisions your team already made — so they stop rewriting what you already decided against.

Who is dlogs built for?

Staff and platform engineers (and their EMs) at 10–50 engineer GitHub-native teams using Cursor, Claude Code, or Copilot heavily. Buyers are typically CTOs or founders.

Does dlogs store everything?

No. dlogs is selective by design. It stores key consequential decisions, not every message or artifact — not a wiki.

Does AI write decisions for us?

No. Agents may propose drafts with explicit provenance, but humans approve and own decision accountability by default.

Does dlogs block pull requests?

No. Current GitHub checks are advisory: they surface governing decision context without acting as a merge gate.

How is this different from ADRs in a /decisions folder?

PR-history mining fills the record retroactively. Agents query it semantically over MCP with lineage and provenance. Records are append-only per org — not files that can be silently edited or lost in a repo tree.

Why does immutable history matter?

It protects decision integrity over time. Teams and agents can see what changed, what was superseded, and why a choice existed at a specific moment.

Where do Slack and Linear fit?

They are supporting capture surfaces. The launch story is agent context over MCP, PR-history mining, append-only trust, and advisory GitHub comments.

Start

Give your agents the decisions you already made.

Free during beta. Connect GitHub, confirm mined drafts, and let Cursor or Claude Code query them over MCP.

Start free beta