tony · tony · May 31, 2026 · May 30, 2026 · May 30, 2026 · May 30, 2026
diff --git a/CHANGES b/CHANGES
@@ -42,6 +42,22 @@ $ uvx --from 'agentgrep' --prerelease allow python
 
 <!-- END PLACEHOLDER - ADD NEW CHANGELOG ENTRIES BELOW THIS LINE -->
 
+### What's new
+
+#### OpenCode backend (#30)
+
+agentgrep now searches [OpenCode](https://github.com/anomalyco/opencode)
+(formerly `sst/opencode`). OpenCode keeps conversations in a single
+SQLite database (`opencode.db`) under `~/.local/share/opencode`, so it
+joins the relational `session → message → part` tables and surfaces each
+text-bearing part — user prompts, assistant replies, model reasoning,
+and subtask prompts — with the session title, working directory, and
+model attached. Discovery honours `XDG_DATA_HOME` and the `OPENCODE_DB`
+override, and every other on-disk store (the legacy JSON layout, config,
+snapshots, the repo cache, logs, and tool output) is catalogued for
+completeness, with `auth.json` documented but never indexed. See
+{doc}`/backends/opencode` for details.
+
 ## agentgrep 0.1.0a12 (2026-05-31)
 
 agentgrep 0.1.0a12 adds Pi (the earendil-works "Pi Agent Harness") as

diff --git a/README.md b/README.md
@@ -5,7 +5,7 @@
 [![License: MIT](https://img.shields.io/badge/license-MIT-blue.svg)](LICENSE)
 
 Read-only search for local AI agent prompts and history across Codex,
-Claude Code, Cursor, Gemini, Grok, and Pi.
+Claude Code, Cursor, Gemini, Grok, Pi, and OpenCode.
 
 `agentgrep` provides a CLI and an MCP server over the same discovery + parsing layer:
 

diff --git a/docs/backends/index.md b/docs/backends/index.md
@@ -53,6 +53,12 @@ Grok CLI prompt history, session transcripts, memory, logs, and config.
 Pi (earendil-works) session transcripts, settings, prompts, and managed extensions.
 :::
 
+:::{grid-item-card} OpenCode
+:link: opencode
+:link-type: doc
+OpenCode (anomalyco) SQLite session store, config, snapshots, and caches.
+:::
+
 ::::
 
 ## Coverage levels
@@ -91,4 +97,5 @@ cursor-ide
 gemini
 grok
 pi
+opencode
 ```
diff --git a/docs/backends/opencode.md b/docs/backends/opencode.md
@@ -0,0 +1,63 @@
+(backend-opencode)=
+
+# OpenCode
+
+Base path: `~/.local/share/opencode` (env overrides: `XDG_DATA_HOME`, `OPENCODE_DB`).
+
+`observed_version`: `opencode v1.15.11` (observed 2026-05-30).
+
+OpenCode (anomalyco/opencode) stores conversations in a single SQLite
+database, `opencode.db`, under its XDG data directory
+(`${XDG_DATA_HOME:-~/.local/share}/opencode`). Non-stable install
+channels use `opencode-<channel>.db`, and `OPENCODE_DB` can relocate the
+database (an absolute path is used directly). This makes OpenCode a
+SQLite backend, like Grok's `session_search` and Cursor's `state.vscdb`,
+rather than a JSONL-transcript backend.
+
+## Stores
+
+```{storage:agent} opencode
+```
+
+## Record schema
+
+### opencode.db
+
+A relational `session → message → part` schema (Drizzle). A conversation
+turn is reconstructed by joining a `part` row up to its `message` (for
+the role) and `session` (for the title and working directory).
+
+`session` table — one row per session:
+
+| Column | Type | Description |
+|--------|------|-------------|
+| `id` | TEXT | Session id (primary key) |
+| `project_id` | TEXT | Git remote/root hash, or `global` |
+| `directory` | TEXT | Working directory, stored verbatim |
+| `title` | TEXT | Session title |
+| `time_created` / `time_updated` | INTEGER | Unix milliseconds |
+
+`message` table — `id`, `session_id` (FK), and a `data` JSON column:
+
+```json
+{"role": "assistant", "modelID": "...", "providerID": "...",
+ "time": {"created": 1779999665000}, "path": {"cwd": "..."}}
+```
+
+`part` table — `id`, `message_id` (FK), `session_id`, and a `data` JSON
+column holding one content part. The searchable text lives here:
+
+| Part `type` | Searchable field |
+|-------------|------------------|
+| `text` | `text` (user prompts and assistant replies) |
+| `reasoning` | `text` (model thinking) |
+| `subtask` | `prompt` |
+
+A part's `kind` is derived from the joined message `role` (`user` →
+prompt, otherwise history). Tool, file, snapshot, patch, and step-marker
+parts are metadata and stay outside default search. Message timestamps
+are unix-milliseconds and are normalized to ISO-8601.
+
+The legacy pre-migration layout (one JSON file per session, message, and
+part under `storage/`) is documented but no longer searched — current
+installs migrate it into `opencode.db` on startup.
diff --git a/docs/dev/index.md b/docs/dev/index.md
@@ -16,7 +16,7 @@ Cross-commit `hyperfine` sweeps across HEAD, trunk, ranges, lookback, tags, or e
 :::{grid-item-card} Storage catalogue
 :link: storage-catalog
 :link-type: doc
-On-disk store layouts for Codex, Claude Code, Cursor, Gemini CLI, Grok CLI, and Pi — useful for adapter authors and anyone tracing why a record was or wasn't found.
+On-disk store layouts for Codex, Claude Code, Cursor, Gemini CLI, Grok CLI, Pi, and OpenCode — useful for adapter authors and anyone tracing why a record was or wasn't found.
 :::
 
 :::{grid-item-card} Architecture decisions

diff --git a/docs/dev/storage-catalog.md b/docs/dev/storage-catalog.md
@@ -307,6 +307,34 @@ Documentary-only entries cover settings, auth (private credentials),
 models, themes, tools, managed binaries, prompt templates, the debug
 log, and the npm extension install root.
 
+### OpenCode
+
+`observed_version`: ``opencode v1.15.11`` (observed 2026-05-30).
+
+OpenCode (anomalyco/opencode) stores conversations in a single SQLite
+database under `${XDG_DATA_HOME or ${HOME}/.local/share}/opencode/`,
+unlike the JSONL-transcript backends:
+
+- `opencode.db_sqlite.v1` parses the `opencode.db` SQLite database. It
+  joins the relational `part → message → session` tables: each
+  text-bearing `part` row becomes a record whose `kind` comes from the
+  joined message `role` (`user` → prompt, else history). Searchable text
+  is `part.data` of type `text`/`reasoning` (the `text` field) and
+  `subtask` (the `prompt`); the session `title`, `directory`, and the
+  message `model`/timestamp are attached. Message times are
+  unix-milliseconds, normalized to ISO-8601.
+
+Discovery resolves the data root via `XDG_DATA_HOME` (default
+`~/.local/share`) plus the `opencode` segment and finds `opencode.db` by
+filename — not a glob, so the binary SQLite file bypasses the text
+prefilter. An absolute `OPENCODE_DB` value is discovered as that exact
+file, so channel installs are reachable by pointing `OPENCODE_DB` at
+their `opencode-<channel>.db`.
+
+Documentary-only entries cover the legacy per-file JSON layout, config,
+auth (private credentials), snapshots, the repo cache, logs, and tool
+output.
+
 ## Adding or updating a store
 
 1. Edit `src/agentgrep/store_catalog.py`. Stamp `observed_version`

diff --git a/docs/getting-started/configuration.md b/docs/getting-started/configuration.md
@@ -12,7 +12,7 @@ Use `--agent` one or more times to limit search or discovery:
 $ uv run agentgrep grep "cache" --agent codex
 ```
 
-Supported agents are `codex`, `claude`, `cursor-cli`, `cursor-ide`, `gemini`, `grok`, and `pi`. Omitting `--agent` searches all supported agents.
+Supported agents are `codex`, `claude`, `cursor-cli`, `cursor-ide`, `gemini`, `grok`, `pi`, and `opencode`. Omitting `--agent` searches all supported agents.
 
 ## Search type
 

diff --git a/docs/index.md b/docs/index.md
@@ -2,7 +2,7 @@
 
 # agentgrep
 
-Read-only search for local AI agent prompts and history across Codex, Claude Code, Cursor, Gemini, Grok, and Pi.
+Read-only search for local AI agent prompts and history across Codex, Claude Code, Cursor, Gemini, Grok, Pi, and OpenCode.
 
 ```{warning}
 **Pre-alpha.** APIs may change. [Feedback welcome](https://github.com/tony/agentgrep/issues).

diff --git a/docs/mcp/resources.md b/docs/mcp/resources.md
@@ -26,7 +26,7 @@ used to interpret that source.
 ```{fastmcp-resource-template} agentgrep_sources_by_agent
 ```
 
-Read `agentgrep://sources/codex`, `agentgrep://sources/claude`, `agentgrep://sources/cursor-cli`, `agentgrep://sources/cursor-ide`, `agentgrep://sources/gemini`, `agentgrep://sources/grok`, or `agentgrep://sources/pi` to filter discovery by agent.
+Read `agentgrep://sources/codex`, `agentgrep://sources/claude`, `agentgrep://sources/cursor-cli`, `agentgrep://sources/cursor-ide`, `agentgrep://sources/gemini`, `agentgrep://sources/grok`, `agentgrep://sources/pi`, or `agentgrep://sources/opencode` to filter discovery by agent.
 
 ## Store catalog
 

diff --git a/docs/tui/index.md b/docs/tui/index.md
@@ -3,7 +3,7 @@
 # TUI
 
 The `agentgrep ui` command launches the interactive Textual explorer
-over the same Codex, Claude Code, Cursor, Gemini, Grok, and Pi stores the rest
+over the same Codex, Claude Code, Cursor, Gemini, Grok, Pi, and OpenCode stores the rest
 of the CLI walks. It is read-only — agentgrep never mutates the
 source stores. Bare `agentgrep` prints the directory of choices, so
 the explorer always needs the explicit `ui` subcommand.

diff --git a/pyproject.toml b/pyproject.toml
@@ -1,7 +1,7 @@
 [project]
 name = "agentgrep"
 version = "0.1.0a12"
-description = "Read-only search for local AI agent prompts and history (Codex, Claude Code, Cursor, Gemini, Grok, Pi)"
+description = "Read-only search for local AI agent prompts and history (Codex, Claude Code, Cursor, Gemini, Grok, Pi, OpenCode)"
 requires-python = ">=3.14,<4.0"
 authors = [
   {name = "Tony Narlock", email = "tony@git-pull.com"}
@@ -21,7 +21,7 @@ classifiers = [
   "Typing :: Typed",
 ]
 
-keywords = ["ai", "codex", "claude", "cursor", "gemini", "grok", "pi", "mcp", "search", "agent-history"]
+keywords = ["ai", "codex", "claude", "cursor", "gemini", "grok", "pi", "opencode", "mcp", "search", "agent-history"]
 readme = "README.md"
 packages = [
   { include = "*", from = "src" },