Selvedge

[selvedge]

selvedge.sh  ·  DevHunt  ·  PyPI  ·  GitHub

[Tests] [PyPI] [License: MIT]

Persistent decision memory for AI agents.

Save why you chose an approach, what you rejected, and what would change your mind. Retrieve that context in the next coding session, using a function, database column, API route or dependency as the lookup key.

Selvedge is for anyone using a compatible agent. Connect through local MCP over stdio, or use the CLI from an agent with shell access. Decisions stay in the same SQLite store in .selvedge/ when you switch tools. No hosted account is required; core storage and retrieval make no LLM calls.

Get started · Documentation · Agent compatibility

uv tool install --upgrade selvedgeselvedge demo

Requires uv and Python. Python 3.11–3.14 is recommended; Python 3.10 remains compatible but is end of life. The demo uses a temporary database. Follow the quickstart to connect your agent to a real project.

Use selvedge demo --mcp to save through a local stdio MCP server, restart it and retrieve the same record. Add --json for the checked IDs and versions. This tests the Python installation running Selvedge with disposable data; your agent's configuration, activation and hooks still need a separate check.

Who Selvedge is for

• Solo developers returning to a project after the original conversation is gone.
• Teams that need to pass recorded decisions between sessions and coding tools.

The problem

Code shows the chosen approach. It may leave out the constraint that shaped it or the alternative you ruled out. Selvedge gives those explanations a place in the project and tools to retrieve them before the next edit.

Only explicitly saved decisions can be recalled. A saved explanation is an account supplied by a person or agent; Selvedge does not verify that it is true or guarantee that an agent will consult it. Verify one decision across two sessions to test your setup.

What's new in v0.3.18

Read the recorded reasons before routine review metadata. The optional GitHub review Action puts rejected and reverted approaches, revisions and constraints first. Its comment defaults to a 6,000-character budget, counts omitted events and links to the trusted base store and longer job summary. Unchanged reruns leave the comment untouched.

Healthy runs with no matching history keep the job summary without creating a new comment; an existing comment is refreshed and broken chains remain visible. Missing history is unknown, not approval. The Action still uses only publication-approved base history and never executes PR-head code.

A regression check also protects the library's local save/reopen/retrieve path with site packages unavailable and socket access blocked. No new runtime dependencies, migrations, MCP tools or model calls. Python 3.10 compatibility is retained; use supported Python 3.11–3.14 for new installations. Native Action execution and review usefulness remain unverified.

What's new in v0.3.17

Check recorded decisions before a review. Generate a Markdown digest, then run selvedge export --format markdown -o DECISIONS.md --check to catch drift. An identical digest exits 0 silently; a stale or missing digest exits 1 with a unified diff and regeneration guidance. Check mode preserves the digest and stored records. Review exported reasoning before publishing it.

selvedge demo --mcp saves a synthetic rejection through a real local MCP server, restarts it and retrieves the same record. It uses disposable state; it does not verify your agent's configuration or hook activation.

CLI output now preserves literal brackets in user data, and blame and Markdown exports distinguish rejected approaches from changes that were tried and reverted. Python 3.10 compatibility remains for this patch, but that runtime is now upstream end of life; use tested, supported Python 3.11–3.14 for new installations. No new runtime dependencies, migrations or MCP tools.

Where Selvedge fits

[An agent saves a decision through MCP or CLI. Selvedge stores its reason and rejected alternatives locally. A later session retrieves the record before editing. Git, code review and observability keep their existing roles.]

1. Save a decision. You or your coding agent record a choice, its reason, and any rejected approaches through MCP or the CLI.
2. Keep it with the project. Selvedge stores the record in local SQLite, keyed to the function, column, route or other entity it concerns.
3. Recall it before the next edit. A later session queries the same store and checks whether the earlier constraint still applies.

Git keeps the code history; review tools check changes; observability shows runtime behavior. Selvedge adds the recorded reasons behind decisions. How it works · Agent Trace interchange

How Selvedge compares

Choose the simplest memory mechanism that fits your workflow:

These approaches can be used together. Read the source-linked comparison and instructions versus ADRs versus decision memory for specific selection criteria and limitations.

Why entity history matters. users.email, env/STRIPE_SECRET_KEY, api/v1/checkout and deps/stripe remain useful query targets when their code moves. Explicit rejection and superseding records let a later session inspect what was tried and whether the original constraint still applies.

Why capture time matters. Recording a reason while the context is available can preserve information a diff does not contain. It does not make the reason infallible. Retrieval uses stored records; the coding agent remains responsible for checking applicability and testing its change.

Why changesets matter. Tag events with a shared changeset such as add-stripe-billing to retrieve a task's decisions across tables, environment variables, routes and functions.

Agent Trace interchange. selvedge export --format agent-trace and selvedge import --format agent-trace exchange Agent Trace v0.1.0 records. Reasoning and entity history travel in the dev.selvedge metadata. The mapping and limits document the format; Selvedge vendors the schema and has no runtime dependency on the upstream project.

Quickstart

Connect your agent

With uv installed:

uv tool install --upgrade selvedgeselvedge democd your-projectselvedge setup

Setup detects supported clients. To select a preset explicitly, use --agent with claude-code, codex, copilot, cursor, gemini or windsurf; repeat the flag for multiple clients. These presets are conveniences, not a compatibility limit. For another agent with local stdio MCP support, follow manual setup. Agents with shell access can use the CLI. Prefer pip? Use python -m pip install --upgrade selvedge in a virtual environment. The selvedge-server executable must be on your editor's PATH; launch the editor from that environment or use the executable's absolute path in its MCP config.

Setup asks before changing files, backs up existing content, installs MCP and agent instructions, initializes the project, and offers a Git post-commit hook. Configuration files and lifecycle hooks vary by client. See the compatibility guide for preset paths, activation requirements and limitations.

Restart your agent in the project and approve Selvedge's tools if prompted. Follow your client's project-trust requirements. Ask the agent:

Use Selvedge to record one real approach we considered and rejected in this project. Include the entity, why, and what would change our mind. Do not invent a decision. Show the saved entity path and record ID.

Start a new session and look up the same entity to verify that the decision carries forward. Follow the first-decision verification guide and cross-agent handoff guide for observable checks. MCP access does not automatically capture every decision: the installed instructions guide the agent to use it. Native lifecycle adapters cover supported client events; startup delivery, watched-edit checks and compaction notifications vary by harness. See the agent hook guide before enabling them.

For CI bootstrap or devcontainer.json postCreateCommand:

selvedge setup --non-interactive --yes

Verify the wiring — open a second terminal in the same project:

selvedge watch

Make any change in your AI tool — add a column, rename a function, add an env var. selvedge watch should print the new event within a second of the agent calling log_change. If nothing arrives, run selvedge doctor for a single-command health check that tells you which step is silently broken.

Query your history:

selvedge status                        # recent activity + missing-commit countselvedge diff users                    # all changes to the users tableselvedge diff users.email              # changes to a specific columnselvedge blame payments.amount         # what changed last and whyselvedge history --since 30d           # last 30 days of changesselvedge history --since 15m           # last 15 minutes ('m' = minutes)selvedge changeset add-stripe-billing  # all events for a feature/taskselvedge search "stripe"               # full-text searchselvedge stats                         # observed tool calls and explanation qualityselvedge import migrations/            # backfill from migration filesselvedge export --format csv           # dump history to CSV

cd your-projectselvedge init

selvedge prompt --install path/to/agent-instructions.md

selvedge prompt

selvedge install-hook

Works with compatible agents

Selvedge has no required agent brand or model provider. Its MCP interface uses local stdio: a compatible client must be able to launch selvedge-server and call its tools. Remote-HTTP-only clients need a separate stdio bridge; Selvedge does not provide a hosted endpoint. An agent with shell access can instead use selvedge log and the read commands.

The integrations below are examples. Choose the one matching your client, or configure another stdio MCP client using its own settings format:

/plugin marketplace add masondelan/selvedge/plugin install selvedge@selvedge

claude mcp add selvedge -- selvedge-server

{  "mcpServers": {    "selvedge": { "command": "selvedge-server" }  }}

{  "mcpServers": {    "selvedge": { "command": "selvedge-server" }  }}

{  "mcpServers": {    "selvedge": { "command": "selvedge-server" }  }}

[mcp_servers.selvedge]command = "selvedge-server"

{  "mcpServers": {    "selvedge": { "command": "selvedge-server" }  }}

{  "mcpServers": {    "selvedge": { "command": "selvedge-server" }  }}

How it works

Selvedge runs as a local MCP server or CLI. Compatible agents call its tools as they work, logging structured change events to a local SQLite database.

Each event records:

• What changed (entity path, change type, diff)
• When (timestamp)
• Who (agent, session ID)
• Why (reasoning — captured from the agent's context in the moment)
• Where (git commit, project)

The diff is git's job. The why is Selvedge's.

Selvedge tracks its own history

This repo dogfoods Selvedge: its .selvedge/selvedge.db is committed, so a fresh clone ships with Selvedge's own why-history. Clone it and ask why any part of Selvedge changed:

git clone https://github.com/masondelan/selvedgecd selvedgeselvedge status                       # recent changes to Selvedge itselfselvedge search "telemetry"           # why the opt-in heartbeat shippedselvedge blame selvedge/semantic.py   # why semantic search was added

Every event was logged by the agents that built Selvedge — the same log_change calls this README asks you to make in your own project.

Entity path conventions

users.email           DB column (table.column)users                 DB tablesrc/auth.py::login    Function in a file (path::symbol)src/auth.py           Fileapi/v1/users          API routedeps/stripe           Dependencyenv/STRIPE_SECRET_KEY Environment variable

Prefix queries work everywhere: users returns users, users.email, users.created_at, and any other entity under the users. namespace.

MCP tools

When connected as an MCP server, Selvedge exposes:

CLI reference

selvedge init [--path PATH]               Initialize in projectselvedge status                           Recent activity summaryselvedge diff ENTITY [--limit N]          Change history for entityselvedge blame ENTITY                     Most recent change + contextselvedge history [--since SINCE]          Browse all history              [--entity ENTITY]              [--project PROJECT]              [--changeset CS]              [--summarize]              [--limit N]selvedge changeset [CHANGESET_ID]         Show events in a changeset                  [--list]                or list all changesets                  [--project NAME]                  [--since SINCE]selvedge search QUERY [--limit N]         Full-text searchselvedge prior-attempts ENTITY            Prior attempts + inferred outcome,                       [--description T]   with the tried → reverted →                       [--all]             re-opened trail + status line                       [--window 7d]       (--all widens recall)                       [--fuzzy TEXT]      add semantic matches (needs the                                           semantic extra; substring fallback)selvedge supersede ENTITY                 Re-open a reverted decision —                  --reasoning TEXT         append-only, links the prior                  [--constraint TEXT]      reverted event (or --supersedes ID)                  [--stale-when TEXT]                  [--supersedes ID]selvedge index [--model NAME]             Build/update the optional semantic              [--json]                     embeddings index (selvedge[semantic])selvedge stale [--entity ENTITY]          Decisions due for a revisit: past              [--project NAME]            revisit_after + still in use, or              [--agent NAME]              stale_when matched by a later change              [--json]                    ("review suggested")selvedge stats [--since SINCE]            Tool call coverage report (per-tool, per-agent)selvedge doctor [--agent CLIENT] [--json] Health check; optional project agent-hook diagnosticsselvedge install-hook [--path PATH]       Install git post-commit hook                     [--window MIN]       (default 60 minutes)selvedge backfill-commit --hash HASH      Backfill git_commit on recent events                        [--window MIN]    (default 60 minutes)selvedge import PATH                      Import migrations (SQL / Alembic) or              [--format auto|sql|         an Agent Trace file (agent-trace)                 alembic|agent-trace]              [--from-git]                or walk git history for reverts:              [--since REF|DATE]          revert-message commits + deletions              [--project NAME]            become change_type="revert" events              [--dry-run]                 (idempotent on commit + entity)selvedge export [--format json|csv|       Export history (agent-trace =                 markdown|agent-trace]      Agent Trace v0.1.0 records;                                            markdown = reviewable digest)              [--since SINCE]              [--entity ENTITY]              [--ndjson]                  agent-trace: one record per line              [--collapse-by-session]     agent-trace: merge a session into one              [--output FILE]selvedge log ENTITY CHANGE_TYPE           Manually log a change             [--diff TEXT]                CHANGE_TYPE: add, remove, modify,             [--reasoning TEXT]           rename, retype, create, delete,             [--agent NAME]               index_add, index_remove, migrate,             [--commit HASH]              revert, supersede             [--project NAME]             [--changeset CS]             [--revisit-after WHEN]       ISO date or offset (e.g. 90d)             [--rename-from OLD]          OLD path when CHANGE_TYPE is 'rename'             [--constraint TEXT]          the principle behind the decision             [--stale-when TEXT]          what would invalidate it             [--supersedes ID]            with CHANGE_TYPE 'supersede'selvedge migrate-paths                    Re-canonicalize stored entity paths                      [--apply]           (dry-run by default; --apply writes)                      [--json]

All read commands support --json for machine-readable output.

Relative time in --since:

• 15m → last 15 minutes (m = minutes)
• 24h → last 24 hours
• 7d → last 7 days
• 5mo → last 5 months (mo or mon = months)
• 1y → last year

Unparseable inputs (e.g. --since yesterday) exit with a clear error rather than silently returning empty results. ISO 8601 timestamps are also accepted and normalized to UTC.

Configuration

.selvedge/config.toml

Every key is optional; a missing file means the defaults below. Precedence is CLI flag → env var → project .selvedge/config.toml → global ~/.selvedge/config.toml → default. SELVEDGE_DB is the one exception: it always wins for database resolution, because the config file is found by resolving that path. selvedge doctor prints the effective value and the step that produced it for every setting.

retention_days_events     = 0       # 0 = never delete events (the default)retention_days_tool_calls = 90      # local telemetry retentionbackup_keep_last          = 7diff_bytes                = 65536   # truncate oversized diffs at log timereasoning_bytes           = 32768   # truncate oversized reasoningdb_size_warn_mb           = 500     # doctor warns above thisstale_days                = 0       # 0 = offdigest_max_bytes          = 4096    # cap on the session-start digestredaction_patterns        = []      # extra secret shapes to warn about
[hook]watch_globs = ["**/migrations/**", "db/**/*.sql"]

Every key also has an env override (SELVEDGE_DIFF_BYTES, SELVEDGE_RETENTION_DAYS_EVENTS, …).

Reviewing captured intent in a pull request

.selvedge/selvedge.db is a SQLite file, so the reasoning inside it doesn't show up in a diff. If you want a reviewable copy, export a Markdown digest:

selvedge export --format markdown -o .selvedge/DECISIONS.mdgit add .selvedge/DECISIONS.md

The digest is grouped by entity with reverted and rejected decisions first. Explicit reject events appear under Considered and rejected, separate from Tried and reverted. The latest event determines the section; a later supersede moves either kind to Re-opened while retaining the full history. Human-readable blame also labels explicit rejections as "rejected". For API compatibility, JSON status and trail phase keep the existing reverted negative-verdict category; change_type: "reject" identifies a rejection.

The digest is deterministic — regenerating with no new events produces a zero-line diff, so it stays reviewable instead of becoming noise everyone learns to skip. Heading anchors derive from the entity path, so links into it keep working as it grows. Regenerate it in the same commit as the code, or from a pre-commit hook.

To catch a stale or missing digest in a hook or CI, use the same filters and output path with --check:

selvedge export --format markdown -o .selvedge/DECISIONS.md --check

The check exits 0 silently when the UTF-8, LF-terminated bytes match, 1 with a unified diff when stale or missing (or an error when unreadable), and 2 for invalid arguments. It requires Markdown and an explicit file; stdout and other formats are rejected. It does not rewrite the digest, create or migrate a database, change stored records, or start background update and telemetry checks. Use an existing initialized store in CI, selected through the normal database resolution or SELVEDGE_DB. SQLite may maintain its normal WAL locking files while reading. Regenerate without --check to update the digest.

Publishing either the digest or database is optional. Review captured reasoning and metadata for secrets and private information before committing or sharing them; a generated diff can also expose that information in CI logs.

Coverage checking

Wondering how often your agent actually calls log_change? Two ways to check:

# Quick summary in the terminalselvedge stats
# Cross-reference against git commitspython scripts/coverage_check.py --since 30d

The coverage script compares your git log against Selvedge events and shows which commits have associated change events. Low coverage usually means the system prompt needs strengthening — see docs/fallbacks.md for guidance.

In CI (GitHub Action)

The same check ships as the Selvedge Coverage Check composite Action, so you can track agent coverage on every push — and optionally fail the build when it drops:

# .github/workflows/selvedge-coverage.ymlname: Selvedge coverageon: [push, pull_request]jobs:  coverage:    runs-on: ubuntu-latest    steps:      - uses: actions/checkout@v4        with:          fetch-depth: 0            # full history so commits can be matched      - uses: masondelan/[email protected]   # pin to a release tag (or @main for latest)        with:          since: 30d          fail-under: "0.5"         # optional: fail below 50% coverage; omit to report only

It writes a coverage summary to the job summary and exposes coverage-ratio, covered, and total as step outputs. The action cross-references your git history against the Selvedge event log, so the runner needs the project's .selvedge/selvedge.db (commit it, or restore it before this step) and full git history (fetch-depth: 0). Inputs: since, window, limit, fail-under, selvedge-version, python-version, working-directory, db-path.

Contributing

Read the feedback and review process for reporting problems, evaluating feature requests and following up on discussions.

git clone https://github.com/masondelan/selvedgecd selvedgepip install -e ".[dev]"pytest

See CLAUDE.md for architecture details and the phase roadmap.

License

MIT — see LICENSE.

Decision context for reviewers

The shared ledger and PR review Action shows recorded reasons, rejected paths, actor/session attribution and integrity for touched entities. The matched injection pilot reports every trial and limits its conclusion to the tested synthetic tasks.