22 KiB

Raw Blame History

ollie Architecture

Philosophy

ollie's design philosophy is Emacs: a small, extensible core. The Go runtime (agent.Agent) defines what an agent is — an event loop, a backend connection, a message history, and three built-in primitives (shell, tool registry, skill registry). Everything else — orchestration, scheduling, workflows, UIs — is pushed out to the surrounding environment. The primary integration philosophy is Plan 9's "everything is a file." olliesrv exposes agent state and behaviors as files in a 9P namespace. Any program that can read and write files can drive an agent: shell scripts, editors, web apps, cron, containers. For desktop-native integration, olliesrv embeds a D-Bus adapter (org.ollie.SessionManager) alongside its 9P interface. The KDE GUI, plasmoid, and web UI connect this way. For D-Bus-only deployments, -no9p disables the 9P listener entirely. Design principles:

Small extensible core. The agent runtime is minimal; capabilities come from composing external scripts.
Three built-in primitives. shell, tool registry (tool_load/tool_list/tool_active), and skill registry (skill_load/skill_list/skill_active) are the only tools compiled into the core. Everything else is a script.
Two integration paths. 9P filesystem (canonical) and/or D-Bus (conventional). Same core, different surfaces, selectable via flags.
No framework lock-in. Frontends are decoupled via whichever interface they prefer; the core doesn't know or care.

Repository Structure

Single Go module (ollie) with two Git submodules for decoupled frontends:

ollie/
├── agent/           Agent loop, history, hooks, prompt resolution, commands
├── backend/         LLM providers (Anthropic, OpenAI, Ollama, Gemini, Copilot, CodeWhisperer)
├── toolsrv/         Tool server: sandboxed execution, tool registry, skill management
├── session/         Session lifecycle (config, creation, persistence)
├── fs/session/      9P filesystem tree for sessions
├── dbus/            D-Bus adapter (org.ollie.SessionManager)
├── detach/          Background process management (ring buffer, signal)
├── elevate/         Elevation broker (privilege escalation daemon)
├── sandbox/         Landlock sandbox config YAML
├── env/             Environment variable loading
├── log/             Structured logging
├── paths/           XDG path resolution
├── mount/           9P FUSE mount client
├── cmd/             Binaries:
│   ├── olliesrv/        9P server
│   ├── ollie-9p/        9P client
│   └── ollie-remote/    Remote execution server
├── kde/             KDE integration (submodule) — plasmoid, GUI, Kate plugin, KRunner, tray
├── agents/          Agent config JSONs (default, coding, orchestrator, worker, ...)
├── prompts/         Prompt templates (markdown)
├── tools/           Tool scripts (file_read, lsp_*, memory_*, subagent_*, ...)
├── skills/          Domain knowledge files (markdown)
├── scripts/         User-facing scripts organized by namespace (s/, u/, x/)
├── sandbox/         Sandbox config YAML files
└── doc/             Documentation

System Overview

flowchart TB
    subgraph Frontends["Frontends"]
        TUI["s/sh (terminal)"]
        ACME["acme (Plan 9)"]
        ELLIE["ellie (Emacs)"]
        KDE["KDE GUI / Plasmoid"]
    end
    subgraph Integration["Integration Layer"]
        direction LR
        P9["9P Filesystem"]
        DBUS["org.ollie.SessionManager (D-Bus)"]
    end
    subgraph Server["olliesrv"]
        direction TB
        NS["9P Namespace\nsession/ · backends · models"]
        DBA["Embedded D-Bus Adapter"]
    end
    subgraph Core["Agent Engine (per session)"]
        LOOP["Agent Loop\n(agent/loop.go)"]
        SESS["Session State\n(agent/history.go)"]
        TRV["toolsrv.Server\n· shell execution\n· tool registry\n· landlock sandbox"]
    end
    subgraph Backends["LLM Backends"]
        OLLAMA["Ollama"]
        OPENAI["OpenAI / OpenRouter"]
        ANTHROPIC["Anthropic"]
        GEMINI["Gemini"]
        CW["CodeWhisperer"]
    end
    TUI --> P9
    ACME --> P9
    ELLIE --> P9
    KDE --> DBUS
    WEB --> DBUS
    P9 --> NS
    DBUS --> DBA
    NS --> LOOP
    DBA --> LOOP
    LOOP <--> SESS
    LOOP --> TRV
    LOOP <--> Backends
    TRV --> Shell["sandboxed shell\n(landrun)"]
    TRV --> Tools["tool scripts\n(OLLIE_TOOLS_PATH)"]

Core Library

The core is a single Go module (ollie) with no binary. Binaries live in cmd/.

Package Layout

Package	Purpose
`agent/`	Agent struct, loop, history, compaction, hooks, commands, prompt resolution, state
`backend/`	Backend interface + LLM providers (Anthropic, OpenAI, Ollama, Gemini, Copilot, CodeWhisperer)
`toolsrv/`	Tool server: sandboxed shell execution, named tool dispatch, skill management, result tiering, remote execution
`session/`	Session lifecycle, config, persistence
`detach/`	Background process management (ring buffer, signal)
`elevate/`	Elevation broker (privilege escalation daemon, policy)
`sandbox/`	Landrun sandbox configuration and command wrapping
`env/`	Session environment helpers
`log/`	Structured logging
`paths/`	XDG path resolution

Key Types

agent.Agent — the concrete struct frontends drive:

type Agent struct {
    // unexported fields — no public interface
}
func New(cfg AgentConfig) *Agent
func (a *Agent) Submit(ctx context.Context, input string)
func (a *Agent) Interrupt(cause error) bool
func (a *Agent) Queue(prompt string)
func (a *Agent) State() string
func (a *Agent) WaitChange(ctx context.Context, field, current string) (string, bool)
func (a *Agent) Close()
// ... model/backend info, CWD, usage, cost, etc.

backend.Backend — LLM provider contract:

type Backend interface {
    ChatStream(ctx context.Context, messages []Message, tools []Tool, params GenerationParams) (<-chan StreamEvent, error)
    Name() string
    Model() string
    SetModel(model string)
    ContextLength(ctx context.Context) int
    Models(ctx context.Context) []string
}

toolsrv.Runner — minimal tool execution interface:

type Runner interface {
    ListTools() ([]ToolInfo, error)
    CallTool(ctx context.Context, tool string, args json.RawMessage) (json.RawMessage, error)
}

toolsrv.Server — the concrete tool server (implements Runner):

type Server struct { /* ... */ }

Agent Loop (`agent/loop.go`)

The loop is the heart of the system. On each turn:

Stream — call backend.ChatStream() with messages + tool definitions
Dispatch — if the model returns tool calls, execute them via toolsrv.Server
Update — append assistant message + tool results to session history
Repeat — loop until the model produces a final text response (no tool calls) Safety mechanisms:

Consecutive error limits: soft nudge at 5, hard abort at 10
Replan gate: after 8 rounds without a PLAN: block, inject a replan nudge
Stall detection: 5 rounds with unchanged LastAction triggers a nudge
Max steps: configurable per-agent; soft exit when reached
Transient retry: up to 3 retries with exponential backoff for rate limits and 5xx
Context overflow: triggers automatic compaction and retry

Session & Context Management (`agent/history.go`)

History owns the message history as a flat []backend.Message slice. Key behaviors:

TaskState: structured JSON overlay (objective, plan_step, constraints, last_action, next_decision) injected at the top of every turn
Compaction: three-zone strategy when context reaches 50% of model limit:
- Cold zone: structured task state summary
- Warm zone: one-line decision index (10 messages)
- Hot zone: last 8 messages verbatim
Result tiers: tool results classified as Hot (verbatim), Warm (summarized on compaction), or Cold (immediately collapsed)
Persistence: sessions saved as JSON to {sessionsDir}/{id}.json

Hooks (`agent/hooks.go`)

Lifecycle callbacks executed as shell commands with JSON payload on stdin:

Hook	When	Exit codes
`agentSpawn`	Session creation	0=ok, 2=block
`preTurn`	Before each turn (deprecated)	0=inject stdout, 2=block
`postTurn`	After each turn	0=ok
`preTool`	Before each tool call	0=ok, 2=block execution
`postTool`	After each tool call	0=append, 2=replace result
`preCompact`	Before compaction	0=ok
`postCompact`	After compaction	0=ok
`turnError`	On backend error	0=handled (skip retries)

Prompt Resolution (`agent/prompt_resolver.go`)

Agent configs declare prompts as a JSON array of shell commands. Each command is executed with $OLLIE and $PWD available; stdout is concatenated to form the system prompt. This enables dynamic, composable prompts assembled from template fragments.

9P Server (`cmd/olliesrv/`)

olliesrv is the central daemon. It implements a 9P2000 file server that exposes the entire agent namespace:

Namespace Layout

session/
├── new              write key=value to create session
├── idx              session index (tab-separated)
├── ls               exec to list sessions
├── kill             exec to kill a session
├── sh               interactive shell
├── b                batch one-shot (create, submit, wait, print, kill)
├── bfg              batch foreground (submit to existing, wait, print)
├── bbg              batch background (submit, return path)
├── cleanup          kill idle sessions
├── {sid}/
│   ├── plan         session-scoped markdown checklist
│   ├── env          session environment
│   └── agent/
│       └── {aid}/
│           ├── cfg          session config (key=value)
│           ├── ctl          control commands (stop, kill, compact, model, ...)
│           ├── state        current state (idle/thinking/calling)
│           ├── statewait    blocking read until state changes
│           ├── chat         full chat history (streamable)
│           ├── offset       byte offset after last user prompt
│           ├── prompt       write to submit a prompt
│           ├── prompt.prev  last submitted prompt
│           ├── fifo.in      queue a prompt (FIFO input)
│           ├── fifo.out     pop queued prompt (FIFO output)
│           ├── context      rendered context window
│           ├── systemprompt rendered system prompt
│           ├── usage        token usage stats
│           ├── cost         estimated cost
│           ├── ctxsz        context size
│           ├── models       available models
│           ├── tools        tool management (read all, write name to load)
│           ├── toolsrv.loaded  currently loaded tools
│           ├── toolsrv.rev     revision counter
│           ├── tail         exec helper
│           └── proc/        detached background processes

9P Interaction Model

All agent interaction uses the ollie-9p client, which auto-discovers the server via $NAMESPACE and identifies the caller via $OLLIE_UNAME. The 9P filesystem exposes session state, control, and inter-agent communication as files:

Sessions are directories under session/
Prompts, state, chat history, and config are readable/writable files
Permission enforcement uses 9P user principals

Permission Model (`fs/session/perm.go`)

All permissions are declared in a single registry (Perms). Key design:

prompt is mode 0666 — group+other can write, owner (the agent) cannot
chat is mode 0444 — read-only for everyone
ctl is mode 0666 — anyone can send control commands
plan is mode 0666 — readable by peer agents

Session Management (`fs/session/`)

The *fs.Tree IS the session collection. Package functions manage lifecycle:

NewRoot(cfg) — create the root tree
Lookup(tree, id) — find a session by ID
All(tree) — list all sessions
CreateFromRoot(tree, args) — create a new session
KillFromRoot(tree, id) — kill a session
RenameFromRoot(tree, old, new) — rename a session
Shutdown(tree) — clean shutdown

Embedded D-Bus Adapter

olliesrv includes an embedded D-Bus adapter that claims org.ollie.SessionManager on the session bus. This provides:

The same method/signal interface used by KDE frontends
SessionCreated, SessionKilled, StateChanged, ChatUpdated signals
Lifecycle watchers that bridge session state changes to D-Bus signals The adapter is best-effort: if no session bus is available (e.g., headless/container), olliesrv continues functioning as a pure 9P server.

D-Bus Adapter (`dbus/`)

The D-Bus adapter is embedded in olliesrv (no separate daemon). For D-Bus-only deployments, run olliesrv -no9p. This disables the filesystem listener while keeping the full D-Bus interface. Session persistence uses the same ~/.local/share/ollie/sessions/active/ directory.

KDE Integration (`kde/`)

The KDE submodule provides several Qt/QML components that connect to org.ollie.SessionManager over D-Bus:

Component	Path	Purpose
GUI	`kde/gui/`	Standalone Qt window for chat interaction
Plasmoid	`kde/plasmoid/`	Plasma widget with `OllieDBusClient`
Kate plugin	`kde/kate/`	In-editor chat pane and ghost completion
KRunner	`kde/krunner/`	Quick-launch sessions from KRunner
System tray	`kde/tray/`	Tray icon with session status
All KDE components use `QDBusServiceWatcher` to handle daemon restarts gracefully — recreating the `QDBusInterface` when the service reappears and refreshing the session list.

Tool System

Architecture

The tool system has exactly one built-in server (toolsrv.Server) that exposes three primitives:

Primitive	Purpose
`shell`	Run a bash command in a sandbox
`tool_load`/`tool_list`/`tool_active`	Dynamic tool registry
`skill_load`/`skill_list`/`skill_active`	Dynamic skill registry
All other capabilities are tool scripts — executable files discovered from `OLLIE_TOOLS_PATH` (default: `~/.config/ollie/tools/`).

Tool Scripts

Tool scripts are plain executables with a structured header comment declaring their name, description, parameters, and parallelism annotation:

file_read, file_write, file_edit, file_glob, file_grep — filesystem I/O
lsp_definition, lsp_references, lsp_rename, lsp_symbols, lsp_diagnostics, lsp_hover, lsp_completion — LSP bridge (Python)
memory_remember, memory_recall — persistent memory
reasoning_think — scratchpad for externalized reasoning
subagent_spawn, subagent_generate — sub-agent lifecycle
route — orchestrator dispatch
gui_* — KDE desktop automation (screenshot, windows, clipboard, input, ...) Scripts annotated ollie:parallel read can be fanned out concurrently by the agent loop.

Sandboxing

Every code execution step is wrapped with landrun (Landlock LSM). Configuration is YAML-based:

general:
  best_effort: true
filesystem:
  ro: ["/usr", "/lib", "/etc", "{HOME}/.config/ollie"]
  rox: ["/usr/bin", "/usr/local/bin"]
  rw: ["{CWD}", "{TMPDIR}"]
network:
  enabled: true
  unrestricted: true

Template variables ({CWD}, {HOME}, {XDG_*}, {OLLIE_*}) are expanded at runtime. landrun is mandatory — there is no unsandboxed fallback. Steps marked elevated: true bypass the sandbox via the elevation backend (x/elevate).

LLM Backends (`backend/`)

All backends implement the streaming-only Backend interface:

Backend	Wire Format	Notes
`OllamaBackend`	`/api/chat`	Local models, context length from `/api/show`
`OpenAIBackend`	`/v1/chat/completions`	OpenAI, OpenRouter, any compatible API
`AnthropicBackend`	`/v1/messages`	Extended thinking, prompt caching
`GeminiBackend`	Gemini API	Google Gemini models
`CopilotBackend`	GitHub Copilot API	Copilot Chat
`CodeWhispererBackend`	CodeWhisperer API	AWS CodeWhisperer
Shared infrastructure:

streamRequest() — common HTTP streaming with error classification
RateLimitError, TransientError, ContextOverflowError, ToolUnsupportedError — typed errors for loop control
GenerationParams — unified sampling parameters across all backends

Multi-Agent System

Multi-agent coordination is built entirely on the filesystem primitive:

Mechanism

Spawn: write session spec to session/new → new session directory appears
Prompt: write task to session/{id}/prompt
Observe: read session/{id}/state or block on session/{id}/statewait
Result: child writes to session/{parent_id}/prompt when done

Delegation Tools

Tool	Semantics
`route`	Orchestrator dispatch; selects backend+model for a task
`subagent_spawn`	Raw session creation with full control
`subagent_generate`	JIT agent identity generation

Callback Protocol

[from={session_id}]
STATUS: done|error
SUMMARY: <result>
ARTIFACTS: <paths>

The parent never actively waits — results arrive as queued prompts. This is the secondary integration path for non-LLM consumers (web UIs, dashboards, external tools) that want a conventional HTTP API without speaking 9P directly.

Frontends

Frontends connect via either 9P (filesystem) or D-Bus, depending on preference:

flowchart LR
    subgraph Core["agent.Agent"]
        AG["Agent Engine"]
    end
    subgraph Surfaces["Integration Surfaces"]
        P9["9P Filesystem\n(session/ namespace)"]
        DB["D-Bus\n(org.ollie.SessionManager)"]
    end
    subgraph Frontends
        SH["s/sh (terminal)"]
        ACME["acme (Plan 9)"]
        EL["ellie (Emacs)"]
        KG["KDE GUI"]
        KP["KDE Plasmoid"]
        KK["Kate Plugin"]
        KR["KRunner"]
        HTTP["curl / scripts"]
    end
    AG --> P9
    AG --> DB
    SH --> P9
    ACME --> P9
    EL --> P9
    HTTP --> P9
    KG --> DB
    KP --> DB
    KK --> DB
    KR --> DB
    WEB --> DB

Frontends are fully decoupled — the core doesn't know which interface they use.

Script Namespaces

User-facing scripts installed to ~/.config/ollie/scripts/{ns}/:

Namespace	Purpose	Key Scripts
`s/`	Session management	`sh` (interactive), `b` (one-shot), `bfg` (blocking), `bbg` (background)
`u/`	Utility compositions	`complete` (code completion), `optimize` (prompt optimization), `cascade`, `escalate`
`x/`	System plugins	`elevate` (privilege escalation), `prime` (prompt template renderer), `freeloader`

Agent Configuration

Agent configs are JSON files in ~/.config/ollie/agents/:

{
  "prompt": [
    "$OLLIE_CFG_PATH/scripts/x/prime SYSTEM_PROMPT",
    "$OLLIE_CFG_PATH/scripts/x/prime session",
    "$OLLIE_CFG_PATH/scripts/x/prime agent-coding",
    "$OLLIE_CFG_PATH/scripts/x/prime tools-file",
    "$OLLIE_CFG_PATH/scripts/x/prime tools-reasoning"
  ],
  "hooks": {
    "agentSpawn": [],
    "turnError": ["$OLLIE_CFG_PATH/scripts/x/freeloader $OLLIE_SESSION_ID"]
  },
  "maxTokens": 8192,
  "temperature": 0.7,
  "maxSteps": 50
}

The prompt array is the key innovation: each element is a shell command whose stdout is concatenated to form the system prompt. This makes prompts composable, dynamic, and environment-aware.

Data Flow

sequenceDiagram
    participant F as Frontend
    participant S as olliesrv
    participant A as agent.Agent
    participant B as LLM Backend
    participant T as toolsrv.Server
    participant SH as sandbox (landrun)
    F->>S: write to session/{id}/prompt
    S->>A: Submit()
    activate A
    loop Agent Loop
        A->>B: ChatStream(messages, tools)
        activate B
        B-->>A: stream events
        deactivate B
        alt has tool calls
            A->>T: Dispatch(tool, args)
            activate T
            T->>SH: sandboxed execution
            SH-->>T: result
            T-->>A: ToolResult
            deactivate T
            A->>A: append to history
        else no tool calls
            A->>A: turn complete
            break
        end
    end
    A-->>S: turn complete
    deactivate A
    S-->>F: readable via session/{id}/chat

Extension Points

What	How
Add an LLM backend	Implement `backend.Backend`
Add a tool	Drop an executable into `OLLIE_TOOLS_PATH`
Add a tool server	Implement `toolsrv.Runner`, register in `NewToolServer`
Add a frontend	Connect via 9P (filesystem) or D-Bus (method calls)
Add a skill	Drop a markdown file into `skills/`
Add an agent persona	Create a JSON config in `agents/`
Customize sandbox	Create a YAML in `~/.config/ollie/sandbox/`

Key Design Decisions

Two integration surfaces — 9P filesystem (canonical, Unix philosophy) and D-Bus (conventional, for GUI/web consumers). Both are thin adapters over the same core.
Three built-in primitives — shell, tool registry, skill registry. Everything else is a script. This keeps the core small and makes all other tools equal citizens.
Mandatory sandboxing — no unsandboxed fallback. Security is not optional.
Streaming-only backends — simplifies the interface; blocking APIs wrap as single-event streams.
Fire-and-forget sub-agents — no active waiting, no polling. Results arrive as filesystem writes.
Prompt-as-code — system prompts are assembled by running shell commands, not by concatenating static strings.
Per-agent principals — 9P permission enforcement prevents self-prompting loops and enforces access boundaries without application-level checks.

22 KiB Raw Blame History