Context engine for AI coding agents

Your AI agent writes technically correct code that quietly breaks your architecture.

Contexer silently captures every decision your AI makes mid-session and replays them at the start of every future session — so your agent never starts blind.

MIT licensed · Self-hosted · No account required · Claude Code native
· The decay problem

AI agents are brilliant at pattern matching.
Terrible at remembering yours.

01

Documentation rots

READMEs and ADRs lag behind code by weeks. By the time the agent reads them, they're already wrong.

02

Tribal knowledge stays tribal

The reason you don't use that ORM is in someone's head. The agent will helpfully reintroduce it.

03

Rules drift every commit

A new top-level dir, a swapped dependency, a renamed pattern — your CLAUDE.md is stale by Friday.

· How it works

Three steps. No meetings.

01/bootstrap

First run scans your repo

On the first session in a repo, Contexer offers to bootstrap. It scans for CI, infra, and stack signals, then confirms what it inferred before storing anything.

contexer · bootstrap
> review this repo and list bugs

 Contexer: no project context stored
  for this repo. Save decisions so future
  sessions start with full context?
    · yes  — quick (1 question)
    · full — thorough (up to 5)
    · no   — skip

> full

 The scan inferred GitLab CI and
  Terraform. Confirm those?
    1. Yes, both correct
    2. Only GitLab CI
    3. Only Terraform
    4. Neither
02/capture

Capture decisions as you work

Tell the agent a rule once. Contexer stores it as a constraint, convention, or architecture note — with the rationale attached.

contexer · capture
> always bump the version in
  pyproject.toml file without fail

  Called contexer (ctrl+o to expand)

 Stored as a constraint. The release
  sequence is always: branch → bump
  pyproject.toml → PR → merge → tag.

✻ Cogitated for 9s
03/resume

Every future session starts loaded

On SessionStart, global rules and constraints are injected automatically. Architecture and patterns load on demand when the agent reaches for them.

contexer · resume
│ Fable 5 is here! Our newest model
  for complex, long-running work
  Included in your plan limits until
  Jun 22, then switch to usage credits.
  SessionStart:resume says: Contexer:
     1 global rule, 16 constraints,
     8 conventions loaded. 11 arch/
     patterns will be loaded on demand.
· What gets stored

Four types of context. Nothing else.

Constraint
Rules that must always apply
✓ Yes
Convention
Team or project standards
✓ Yes
Architecture
Why structural decisions were made
○ On demand
Pattern
Recurring implementation approaches
○ On demand
· Pricing

Start free. No account needed.

Available today

Open Source

Freeforever

Install in 2 minutes. No account. Works with any git repo.

MIT licensed · Self-hosted · No account required

  • Automatic decision capture
  • Session memory across prompts
  • Rationale injection on "why" questions
  • Global conventions store (cross-repo)
  • Works with any git repo
Early access

Teams

Early access

Shared intelligence layer built on top of the open-source engine.

Contact us

Enterprise

Contact us

For platform teams running agents at scale.

· Teams & Enterprise waitlist

The team layer is coming. Get in line.

Prefer the free open-source version? Install it now →