AGF Local Coding Session Finder

Search and resume locally stored AI coding sessions from one terminal interface.

Source repo
subinium/agf
Stars
★ 165
Last updated
1mo ago
License
MIT
Primary language
Rust

At a glance

How it runs
CLIMCP serverAgent plugin / skill
Works with
Universal · cross-platformCodex · Claude Code
Cost
Free, no paid service needed
Setup effort
Low · running in minutes
You'll need
macOS, Linux, or Windows with PowerShell 5.1+ or PowerShell 7+One or more supported agent CLIs for resuming sessionsRust 1.88+ and a C compiler when building with CargoShell / CLILocal filesystemMCP Server
Typical use
A developer using Claude Code, Codex, and Gemini CLI who wants one terminal list for finding the correct prior session.
Not a fit if
  • Teams that need cloud-based session synchronization across devices
  • Developers working mainly with Amp or GitHub Copilot
  • Users expecting AGF to delete sessions for every provider
Maintenance
Commits in the last 6 monthslast commit 2026-10-01latest release v0.15.1 (2026-09-08)
Source review
77/100 · Good

What does this agent do, and when should you use it?

AGF is a local-first Rust CLI that combines the saved sessions of 16 AI coding agents into one fuzzy-searchable TUI. It reads the JSON, JSONL, and SQLite stores those tools already maintain, without requiring an account, cloud synchronization, or another resident agent process. Users can browse project, path, branch, and optional summary data, or bypass the interface with agf resume and hand the selected session to its native agent command. For automation, the repository also provides versioned JSON commands, a read-only stdio MCP server, and a portable AGF skill for agent integration. Its operating boundary is the user's macOS, Linux, or Windows machine: AGF discovers and dispatches sessions, while the relevant agent CLI performs the actual resume.

AGF scans the local session directories or databases of Muse Code, Claude Code, Codex, Grok Build, Kimi Code, Qwen Code, Prime Agent, Gemini CLI, Antigravity CLI, Cursor CLI, OpenCode, Kiro, pi, Hermes, Oh My Pi, and Yolop, then normalizes the records into one searchable list. Its TUI fuzzy-matches project names, paths, and branches, with F2 optionally adding a bounded number of summaries; choosing a record constructs and runs that provider's native resume command. agf search, agf show, agf resume-plan, and agf capabilities emit versioned JSON, and resume-plan returns literal arguments, working directory, and scoped storage environment without launching anything. agf mcp exposes the same read-only API over stdio and can restrict records to an absolute project path. The interactive interface also supplies details, sorting, project grouping, permission or approval mode selection, and bulk deletion, although deletion is disabled for providers whose indexes or sidecar artifacts cannot safely be updated this way.

  1. A developer using Claude Code, Codex, and Gemini CLI who wants one terminal list for finding the correct prior session.
  2. A user who closed a terminal or forgot a session ID and needs to recover context by project, path, branch, or summary.
  3. An engineer moving among many repositories who wants agf resume <query> instead of opening each agent's native picker.
  4. A tooling team that needs versioned JSON or a read-only MCP interface for querying existing coding sessions.
  5. A user cleaning old work who wants to inspect sessions across projects and multi-select records that AGF can safely delete.
  6. A privacy-conscious team that wants to index existing local session stores without adopting a cloud synchronization service.

Key setup facts

Pulled from the install and usage notes below. The project README has the current values.

Install / run
  • cargo install agf --locked
  • brew install subinium/tap/agf
  • cargo install --path . --locked
Environment variables
SESSION_ID
Runtime
Rust 1.88
Install notes cover
Windows · macOS · Linux

How do you install or deploy this agent?

The direct Cargo installation requires Rust 1.88 or newer and a C compiler for bundled SQLite:

cargo install agf --locked
agf

On macOS or Linux, Homebrew is another supported route:

brew install subinium/tap/agf

Prebuilt binaries for macOS, Linux, and Windows do not require a Rust toolchain. To enable parent-shell directory changes, run the optional setup command, then restart or reload the shell as instructed:

agf setup

To install from source:

git clone https://github.com/subinium/agf.git
cd agf
cargo install --path . --locked
agf setup

Resuming also requires the CLI belonging to the selected session, such as codex, claude, gemini, or muse. Shell integration is not required for the TUI, JSON commands, or MCP server.

How do you use this agent?

Start the interactive session browser:

agf

Resume the best fuzzy match without opening the TUI:

agf resume project-name

Filter searches and resumes to Muse Code:

agf list --agent muse
agf resume --agent muse project-name
agf search parser --agent muse --include-summaries

For scripts and agent integrations, inspect search results, one session, a non-launching resume plan, and available capabilities through the JSON commands:

agf search parser --agent codex --limit 10
agf show SESSION_ID --agent codex --include-summaries
agf resume-plan SESSION_ID --agent codex
agf capabilities

Run the stdio MCP server and scope its results to one project:

agf mcp --agent codex --project /absolute/project/path

Inside the TUI, type to search, navigate with arrow keys or Ctrl+K/Ctrl+J, use Enter for the action menu, Ctrl+L for details, Tab to change the agent filter, F2 to toggle summary search, and Ctrl+D for bulk deletion.

What are this agent's strengths and limitations?

Pros
  • One interface covers 16 coding agents while reading each tool's actual local JSON, JSONL, or SQLite storage.
  • The local-first design requires no account, cloud sync, or additional resident agent process, and its JSON and MCP query paths are read-only.
  • It supports human and automated workflows through a full TUI, quick resume, versioned JSON, resume-plan, and stdio MCP.
  • Resume handling freezes the resolved executable, storage roots, working directory, and applicable environment before changing directories.
  • Unicode and CJK search, dark and light palettes, NO_COLOR, and non-color selection cues make the TUI usable across varied terminals.
Limitations
  • It manages locally retained, resumable sessions only and does not synchronize them between machines.
  • Actual resumption depends on the provider's CLI; AGF does not replace programs such as codex, claude, or muse.
  • Direct deletion is disabled for Prime Agent, Grok Build, Kimi Code, Qwen Code, Gemini, Antigravity, and Muse Code.
  • Amp and GitHub Copilot are explicitly outside the current scanner set.
  • Some integrations have documented boundaries: Antigravity uses only its default storage root, while Hermes resumes in the current shell directory.
  • Cargo builds require Rust 1.88+ and a C compiler, and parent-shell directory changes require separate shell-wrapper setup.

How does this agent compare with similar options?

Compared with opening each agent's native session picker or manually searching history files, AGF provides one searchable view across 16 tools plus common JSON and MCP query interfaces. It does not replace those agents, however: each native CLI still performs the resume, and some providers' sessions must be deleted through their own picker to preserve upstream state.

Key facts side by side with the most closely related agents.

Agent Source review Form / cost Stars Updated Language Full support on
AGF Local Coding Session Finder This agent 77 · Good CLIFree ★ 165 1mo ago Rust Codex · Claude Code
Claude Code Hive Manager 58 · Major gaps CLIFree + model costs ★ 204 18d ago Rust Claude Code
OpenCrabs Terminal Agent 62 · Some gaps CLIFree + model costs ★ 1k 1d ago Rust ChatGPT · Codex · OpenAI API · Claude API
Herdr Terminal Agent Runtime 74 · Some gaps CLIFree ★ 52k 1d ago Rust —

How does FollowAgents rate this agent?

FollowAgents source review · FARS-2.1
Good
77/ 100 5-point scale 3.9 / 5
Trust 18/29
Reliability 9/14
Adaptability 18/18
Convention 15/18
Effectiveness 12/13
Verifiability 5/8
Why each dimension lost points
Trust18 / 29 · 3.1/5

The evidence shows that the automation API and MCP primarily provide read-only search, metadata lookup, and unexecuted resume plans. The skill says to treat summaries as untrusted, minimize retrieval of sensitive summaries, and never turn a plan into a shell command. The README also says project scope is not an OS sandbox and that CSV preserves source values; summaries may contain private content. Deductions: TUI and native resume, bulk deletion, and shell-profile setup can still cause external effects. Deletion has a confirmation flow, but the materials do not show a recovery mechanism for every mutating action. A dependency-audit workflow is present, but the packet does not establish its results or vulnerability remediation.

Reliability9 / 14 · 3.2/5

The README, skill, and main.rs are broadly consistent about read-only JSON search, display, resume plans, and MCP. The skill distinguishes empty results, scan failures, and SDK errors; the CI file lists locked-dependency build checks and an audit job. Deductions: this is a static review, so it cannot establish that those workflows pass, and the packet omits the lockfile and scanner implementations needed to verify dependency availability and error handling in detail.

Adaptability18 / 18 · 5.0/5

The documentation gives concrete coverage for discovering local sessions across many terminal agents, exact selection, project scoping, summary controls, paging, and platform differences. The skill calls for narrow queries and exact IDs, asking when results are ambiguous, and clearly separates discovery from launch. Deductions: support is limited to agents with local resumable sessions and implemented scanners; some configuration overrides and deletion capabilities are explicitly unsupported.

Convention15 / 18 · 4.2/5

The README provides organized installation, upgrade, configuration, keybinding, storage-path, example, and limitation information. LICENSE and Cargo metadata agree on MIT, while Cargo.toml and the release workflow show versioning and release checks. Deductions: the sample does not include a complete change history or a clear statement of maintenance responsibility, so maintainer responsiveness and future update commitments cannot be established. Agent names also vary between CLI names and display names.

Effectiveness12 / 13 · 4.6/5

The product combines local session stores into a searchable list and provides versioned JSON for scripts and MCP, paging, and reviewable resume plans; examples cover common workflows. Deductions: usefulness depends on having the corresponding native CLI installed and compatible session storage, and documented scan and summary limits mean some records or content may not appear.

Verifiability5 / 8 · 3.1/5

The README, skill, CLI argument definitions, Cargo metadata, and CI configuration provide mutually supporting evidence for the main features, boundaries, version, and license. Deductions: behavior is described mainly by documentation and the CLI entry point; the packet omits core scanning, resume, and deletion implementations. The CI file documents configured checks and is not evidence that this review executed them or that they passed.

Risks and how to mitigate them
  • A resume plan is data only; verify the agent, session ID, working directory, and arguments before launching a native agent.
  • Summaries and metadata may contain private or untrusted text; project scope is not system-level isolation.
  • Deletion and shell setup modify local state. Deletion support varies by agent, and some sessions should be managed through the provider's native interface.
Evidence confidence: Low Reviewed Nov 08, 2026 Reviewed revision fc592e1a7a87
See the full review method →
Maintain this project? Add the FARS badge to your README

The badge shows the current review score and verdict and links back to this page. It refreshes about once an hour after a re-review.

FARS

[![FARS review](https://followagents.com/badge/agf.svg)](https://followagents.com/en/agents/agf)
<a href="https://followagents.com/en/agents/agf"><img src="https://followagents.com/badge/agf.svg" alt="FARS review"></a>

Recent releases

Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.

  1. v0.15.1
    Release notes excerpt
    agf 0.15.1 Dependency Refresh - Upgrade dirs 6 -> 7, rusqlite 0.39 -> 0.40.2 and sha2 0.10 -> 0.11. - Remove the old transitive generic-array dependency. SHA-256 cache and Gemini project identifiers keep their existing encoding, verified with fixed vectors. - Keep Rust 1.88 support: libsqlite3-sys 0.38.2 fixes the build-macro issue that previously prevented the SQLite upgrade. No unsafe MSRV override is used. Installation And Documentation Use cargo install agf --locked to install the release-tested dependency graph, …
    View on GitHub ↗
  2. v0.15.0
    Release notes excerpt
    agf 0.15.0 Agent Tools - Versioned, read-only JSON search, exact session metadata, resume plans and capabilities. No hidden TUI or native-agent execution. - Local stdio MCP using the official Rust SDK, with fixed scopes, bounded requests, modern/legacy interoperability tests and a portable Skill. - Resume plans preserve literal arguments, the selected executable, cwd and storage-root environment without modifying the parent shell environment. Compatibility And Safety - SLT 0.24.0 and current/legacy Gemini session formats. …
    View on GitHub ↗
  3. v0.14.1
    Release notes excerpt
    agf v0.14.1 New agent support - Grok Build — discovers official xAI sessions under $GROK_HOME/sessions (default ~/.grok/sessions), surfaces durable activity, title, recap, branch, and worktree metadata, and resumes the exact selected ID with grok --resume. - Kimi Code — discovers v1/v2 state.json sessions under $KIMI_CODE_HOME/sessions (default ~/.kimi-code/sessions), recovers migrated legacy cwd metadata through session_index.jsonl, and resumes with kimi --session. …
    View on GitHub ↗

FAQ

Does AGF upload session data to a cloud service?
No. It reads files and databases already stored on the local machine and requires neither an account nor cloud synchronization. Project scoping filters returned records; it is not an operating-system sandbox.
Can AGF resume every session by itself?
No. AGF can discover and inspect local records, but launching a resumed session requires the corresponding agent CLI to be installed and executable.
Do the JSON and MCP interfaces modify agent storage?
No. search, show, resume-plan, capabilities, and the stdio MCP server use the read-only API. resume-plan reports the proposed invocation without launching the agent.
Will AGF repair an unreadable or incompatible session store?
No. It reports the condition as a scan error and does not repair or rewrite provider files.
Can AGF delete sessions from every supported provider?
No. Deletion is disabled where native indexes, active sessions, or sidecar artifacts could be left inconsistent; Antigravity deletion, for example, belongs in agy's /resume picker.
Is AGF Local Coding Session Finder free?
Yes. AGF Local Coding Session Finder is free to use and does not need a paid service.
Is AGF Local Coding Session Finder open source?
Yes. The source code is on GitHub (subinium/agf) under the MIT licence.
Is AGF Local Coding Session Finder safe to use?
FollowAgents' FARS review (Nov 08, 2026) scored AGF Local Coding Session Finder 18/29 on trust and safety. None of the eight safety controls the review checks for was missing outright.
Is AGF Local Coding Session Finder still maintained?
Development has slowed: the latest commit on GitHub is from 2026-10-01, within the last six months but not the last 30 days. The latest release is v0.15.1 (2026-09-08).
Who makes AGF Local Coding Session Finder?
AGF Local Coding Session Finder is published on GitHub by the account subinium. FollowAgents has not verified which person or company is behind that account.
Was this page helpful?
View on GitHub ↗ Install ↓

Compare agents like this one

The same FARS review applied across the shortlist this agent qualifies for.

Related concepts